From mboxrd@z Thu Jan  1 00:00:00 1970
From: Jan Kiszka <jan.kiszka@siemens.com>
Subject: Re: [RFC][PATCH] KVM: SVM: Sync g_pat with guest-written PAT value
Date: Mon, 20 Apr 2015 19:21:53 +0200
Message-ID: <553535B1.3050901@siemens.com>
References: <552B5128.4010909@siemens.com> <552B6923.3020602@siemens.com> <20150420161401.GB26491@potion.brq.redhat.com> <20150420171635.GC26491@potion.brq.redhat.com>
Mime-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: QUOTED-PRINTABLE
Cc: kvm <kvm@vger.kernel.org>, Joel Schopp <joel.schopp@amd.com>
To: =?UTF-8?B?UmFkaW0gS3LEjW3DocWZ?= <rkrcmar@redhat.com>
Return-path: <kvm-owner@vger.kernel.org>
Received: from goliath.siemens.de ([192.35.17.28]:55500 "EHLO
	goliath.siemens.de" rhost-flags-OK-OK-OK-OK) by vger.kernel.org
	with ESMTP id S1752332AbbDTRWE (ORCPT <rfc822;kvm@vger.kernel.org>);
	Mon, 20 Apr 2015 13:22:04 -0400
In-Reply-To: <20150420171635.GC26491@potion.brq.redhat.com>
Sender: kvm-owner@vger.kernel.org
List-ID: <kvm.vger.kernel.org>

On 2015-04-20 19:16, Radim Kr=C4=8Dm=C3=A1=C5=99 wrote:
> 2015-04-20 18:14+0200, Radim Kr=C4=8Dm=C3=A1=C5=99:
>> Tested-by: Radim Kr=C4=8Dm=C3=A1=C5=99 <rkrcmar@redhat.com>
>=20
> Uncached accesses were roughly 20x slower.
> In case anyone wanted to reproduce, I used this as a kvm-unit-test:
>=20
> ---
> #include "processor.h"
>=20
> #define NR_TOP_LOOPS 24
> #define NR_MEM_LOOPS 10
> #define MEM_ELEMENTS 1024
>=20
> static volatile u64 pat_test_memory[MEM_ELEMENTS];
>=20
> static void flush_tlb(void)
> {
> 	write_cr3(read_cr3());
> }
>=20
> static void set_pat(u64 val)
> {
> 	wrmsr(0x277, val);
> 	flush_tlb();
>=20
> }
>=20
> static u64 time_memory_accesses(void)
> {
> 	u64 tsc_before =3D rdtsc();
>=20
> 	for (unsigned loop =3D 0; loop < NR_MEM_LOOPS; loop++)
> 		for (unsigned i =3D 0; i < MEM_ELEMENTS; i++)
> 			pat_test_memory[i]++;
>=20
> 	return rdtsc() - tsc_before;
> }
>=20
> int main(int argc, char **argv)
> {
> 	unsigned error =3D 0;
>=20
> 	for (unsigned loop =3D 0; loop < NR_TOP_LOOPS; loop++) {
> 		u64 time_uc, time_wb;
>=20
> 		set_pat(0);
> 		time_uc =3D time_memory_accesses();
>=20
> 		set_pat(0x0606060606060606ULL);
> 		time_wb =3D time_memory_accesses();
>=20
> 		if (time_uc < time_wb * 4)
> 			error++;
>=20
> 		printf("%02d uc: %10lld wb: %8lld\n", loop, time_uc, time_wb);
> 	}
>=20
> 	report("guest PAT", !error);
>=20
> 	return report_summary();
> }
>=20

Great, thanks. Will you push it to the unit tests? Could raise
motivations to fix the !NPT/EPT case.

Jan

--=20
Siemens AG, Corporate Technology, CT RTC ITP SES-DE
Corporate Competence Center Embedded Linux