Re: [PATCH] x86/clear_user: Make it faster

[Borislav Petkov <bp@alien8.de>]

On Tue, May 24, 2022 at 02:32:36PM +0200, Borislav Petkov wrote:
> I still need to run it on production Icelake and that is kinda being
> delayed due to server room cooling issues (don't ask ;-\).

So I finally got a production level ICL-X:

[    0.220822] smpboot: CPU0: Intel(R) Xeon(R) Gold 6336Y CPU @ 2.40GHz (family: 0x6, model: 0x6a, stepping: 0x6)

and frankly, this looks really weird:

clear_user_original:
Amean: 19679.4 (Sum: 13652560764, samples: 693750)
Amean: 19743.7 (Sum: 13693470604, samples: 693562)

(I ran it twice just to be sure.)

ERMS:
Amean: 20374.3 (Sum: 13910601024, samples: 682752)
Amean: 20453.7 (Sum: 14186223606, samples: 693576)

FSRM:
Amean: 20458.2 (Sum: 13918381386, sample s: 680331)

so either that particular box is weird or Icelake really is slower wrt
FSRM or I've fat-fingered it somewhere.

But my measuring is as trivial as:

static __always_inline unsigned long clear_user(void __user *to, unsigned long n)
{
	if (access_ok(to, n)) {
		unsigned long start, end, ret;

		start = rdtsc_ordered();
		ret = __clear_user(to, n);
		end = rdtsc_ordered();

		trace_printk("to: 0x%lx, size: %ld, cycles: %lu\n", (unsigned long)to, n, end - start);

		return ret;
	}

	return n;
}

so if anything I don't see what the problem could be...

Hmm.

-- 
Regards/Gruss,
    Boris.

https://people.kernel.org/tglx/notes-about-netiquette