Messages in this thread | | | Date | Mon, 6 Dec 2021 13:08:13 -0800 | From | "Paul E. McKenney" <> | Subject | Re: [PATCH 2/5] nptl: Add rseq registration |
| |
On Mon, Dec 06, 2021 at 09:26:51PM +0100, Florian Weimer wrote: > * Paul E. McKenney via Libc-alpha: > > >> The C memory model is broken and does not prevent out-of-thin-air > >> values. As far as I know, this breaks single-copy atomicity. In > >> practice, compilers will not exercise the latitude offered by the memory > >> model. volatile does not ensure absence of data races. > > > > Within the confines of the standard, agreed, use of the volatile keyword > > does not explicitly prevent data races. > > > > However, volatile accesses are (informally) defined to suffice for > > device-driver memory accesses that communicate with devices, whether via > > MMIO or DMA-style shared memory. The device-driver firmware is often > > written in C or C++. So doesn't this informal device-driver guarantee > > need to also do what is needed for userspace code that is communicating > > with kernel code? If not, why not? > > The informal guarantee is probably good enough here, too. However, the > actual accesses are behind macros, and those macros use either > non-volatile plain reads or inline assembler (which use > single-instruction naturally aligned reads).
Agreed, a non-volatile plain read is quite dangerous in this context.
Thanx, Paul
| |