From mboxrd@z Thu Jan 1 00:00:00 1970 From: Ingo Molnar Subject: Re: S3 sleep no longer works on x86_64 in 2.6.27 Date: Thu, 23 Oct 2008 14:51:04 +0200 Message-ID: <20081023125104.GA9506@elte.hu> References: <20081023122405.GC14838@redhat.com> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Return-path: Received: from mx2.mail.elte.hu ([157.181.151.9]:57313 "EHLO mx2.mail.elte.hu" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752414AbYJWMvc (ORCPT ); Thu, 23 Oct 2008 08:51:32 -0400 Content-Disposition: inline In-Reply-To: <20081023122405.GC14838@redhat.com> Sender: linux-next-owner@vger.kernel.org List-ID: To: Gleb Natapov Cc: gcosta@redhat.com, linux-next@vger.kernel.org, "Rafael J. Wysocki" , Thomas Gleixner , "H. Peter Anvin" * Gleb Natapov wrote: > Hi, > > S3 sleep no longer works on x86_64 (at least in KVM, but it looks like > this is the kernel bug). Kernel 2.6.26 works. I think that the commit > that caused the problem is a939098af, but I can't be 100% sure since > compilation is broken at this point. > > Triple fault happens during S3 resume. It happens in > arch/x86/kernel/head_64.S during access to GDT after it was loaded on > line 213 (lgdt early_gdt_descr(%rip)) early_gdt_descr points to > per_cpu__gdt_page and this address contains valid GDT entries during a > regular boot, but on S3 resume in contains garbage. It seems that > per_cpu area is reallocated somewhere, but I don't understand this > magic enough to fix it. Can somebody look at this and explain what > happens? could you check whether the (post-v2.6.27) upstream fix below does the trick for you? Ingo ---------------> >>From 3038edabf48f01421c621cb77a712b446d3a5d67 Mon Sep 17 00:00:00 2001 From: Rafael J. Wysocki Date: Fri, 17 Oct 2008 01:26:27 +0200 Subject: [PATCH] x86 ACPI: fix breakage of resume on 64-bit UP systems with SMP kernel x86 ACPI: Fix breakage of resume on 64-bit UP systems with SMP kernel We are now using per CPU GDT tables in head_64.S and the original early_gdt_descr.address is invalidated after boot by setup_per_cpu_areas(). This breaks resume from suspend to RAM on x86_64 UP systems using SMP kernels, because this part of head_64.S is also executed during the resume and the invalid GDT address causes the system to crash. It doesn't break on 'true' SMP systems, because early_gdt_descr.address is modified every time native_cpu_up() runs. However, during resume it should point to the GDT of the boot CPU rather than to another CPU's GDT. For this reason, during suspend to RAM always make early_gdt_descr.address point to the boot CPU's GDT. This fixes http://bugzilla.kernel.org/show_bug.cgi?id=11568, which is a regression from 2.6.26. Signed-off-by: Rafael J. Wysocki Acked-by: Pavel Machek Cc: Signed-off-by: Ingo Molnar Reported-and-tested-by: Andy Wettstein --- arch/x86/kernel/acpi/sleep.c | 3 +++ 1 files changed, 3 insertions(+), 0 deletions(-) diff --git a/arch/x86/kernel/acpi/sleep.c b/arch/x86/kernel/acpi/sleep.c index 426e5d9..c44cd6d 100644 --- a/arch/x86/kernel/acpi/sleep.c +++ b/arch/x86/kernel/acpi/sleep.c @@ -10,6 +10,7 @@ #include #include #include +#include #include "realmode/wakeup.h" #include "sleep.h" @@ -98,6 +99,8 @@ int acpi_save_state_mem(void) header->trampoline_segment = setup_trampoline() >> 4; #ifdef CONFIG_SMP stack_start.sp = temp_stack + 4096; + early_gdt_descr.address = + (unsigned long)get_cpu_gdt_table(smp_processor_id()); #endif initial_code = (unsigned long)wakeup_long64; saved_magic = 0x123456789abcdef0;