On Fri, 19 Jan 2018, Kirill A. Shutemov wrote:
> This patch implements infrastructure to call WBNOINVD instruction.
grep 'This patch' Documentation/process/
> The basic idea is to match what we have for WBINVD.
Is it just an idea or is it actually doing it?
> The instruction is defined as WBINVD with REP (0xf3) prefix. If the
> machine doesn't support the instruction the prefix will be ignored and
> we would fallback to WBINVD.
We fallback? We do nothing.
The CPU ignores the prefix and falls back to WBINVD, right?
> +
> +#define wbnoinvd_on_cpu(cpu) wbnoinvd()
> +static inline int wbnoinvd_on_all_cpus(void)
Bah. Please stop glueing defines in front of the function body. That's just
makes it harder to read.
Thanks,
tglx
