Centrino Duo not working so hot

Centrino Duo not working so hot

[David Brown]

I remember this bug before and I thought I figured it out by using the
bootloader on the image instead of using the -kernel and -initrd
commandline options to qemu.

However, the guest just stopping has come back. For some reason using
the latest kvm with 2.6.21 both in fedora 7 (kvm-25) and debian
unstable (kvm-17) the guest os just stops after a while. It stops at
random places during the kernel boot and never gets further than
starting udev.  Grub and other etherboot images work fine. Also, using
-no-kvm (however really slow) works fine as well.

I was wondering if anyone else has seen this?

processor       : 1
vendor_id       : GenuineIntel
cpu family      : 6
model           : 14
model name      : Genuine Intel(R) CPU           U2500  @ 1.20GHz
stepping        : 8
cpu MHz         : 1200.000
cache size      : 2048 KB
physical id     : 0
siblings        : 2
core id         : 1
cpu cores       : 2
fdiv_bug        : no
hlt_bug         : no
f00f_bug        : no
coma_bug        : no
fpu             : yes
fpu_exception   : yes
cpuid level     : 10
wp              : yes
flags           : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge
mca cmov pat clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe nx
constant_tsc pni monitor vmx est tm2 xtpr
bogomips        : 2394.71
clflush size    : 64

I remember Avi stating that this is a Core not 2, is this just a
crappy cpu selection on my part (or Dells part)?

Thanks,
- David Brown

-------------------------------------------------------------------------
This SF.net email is sponsored by DB2 Express
Download DB2 Express C - the FREE version of DB2 express and take
control of your XML. No limits. Just data. Click to get it now.
http://sourceforge.net/powerbar/db2/

Re: Centrino Duo not working so hot

[Avi Kivity]

David Brown wrote:
> I remember this bug before and I thought I figured it out by using the
> bootloader on the image instead of using the -kernel and -initrd
> commandline options to qemu.
>
> However, the guest just stopping has come back. For some reason using
> the latest kvm with 2.6.21 both in fedora 7 (kvm-25) and debian
> unstable (kvm-17) the guest os just stops after a while. It stops at
> random places during the kernel boot and never gets further than
> starting udev.  Grub and other etherboot images work fine. Also, using
> -no-kvm (however really slow) works fine as well.
>
> I was wondering if anyone else has seen this?
>   

Does the problem occur with 'taskset 1 qemu ...'?  This pins the guest 
on one core.

>
> I remember Avi stating that this is a Core not 2, is this just a
> crappy cpu selection on my part (or Dells part)?
>   

The core-not-2's main problem is that I don't test regularly with it.  
It is slightly strange (no 64-bit support, hence some important msrs are 
missing) so it has triggered a lot of kvm bugs.

-- 
error compiling committee.c: too many arguments to function


-------------------------------------------------------------------------
This SF.net email is sponsored by DB2 Express
Download DB2 Express C - the FREE version of DB2 express and take
control of your XML. No limits. Just data. Click to get it now.
http://sourceforge.net/powerbar/db2/

Re: Centrino Duo not working so hot

[David Brown]

> Does the problem occur with 'taskset 1 qemu ...'?  This pins the guest
> on one core.

Sorry for the late response but yes that seems to be working fine now.

> The core-not-2's main problem is that I don't test regularly with it.
> It is slightly strange (no 64-bit support, hence some important msrs are
> missing) so it has triggered a lot of kvm bugs.

Ah okay good thing I have one and can test things then ;)

Thanks,
- David Brown

-------------------------------------------------------------------------
This SF.net email is sponsored by DB2 Express
Download DB2 Express C - the FREE version of DB2 express and take
control of your XML. No limits. Just data. Click to get it now.
http://sourceforge.net/powerbar/db2/

Re: Centrino Duo not working so hot

[Avi Kivity]

[-- Attachment #1: Type: text/plain, Size: 699 bytes --]

David Brown wrote:
>> Does the problem occur with 'taskset 1 qemu ...'?  This pins the guest
>> on one core.
>
> Sorry for the late response but yes that seems to be working fine now.
>
>> The core-not-2's main problem is that I don't test regularly with it.
>> It is slightly strange (no 64-bit support, hence some important msrs are
>> missing) so it has triggered a lot of kvm bugs.
>
> Ah okay good thing I have one and can test things then ;)
>

Please check the attached patch (without taskset).  The patch attempts
to keep the time stamp counter monotonic, even if the guest migrates
from core to core.

-- 
Do not meddle in the internals of kernels, for they are subtle and quick to panic.


[-- Attachment #2: vmx-tsc-sync.patch --]
[-- Type: text/x-patch, Size: 806 bytes --]

diff --git a/drivers/kvm/vmx.c b/drivers/kvm/vmx.c
index b909b54..6beb0f1 100644
--- a/drivers/kvm/vmx.c
+++ b/drivers/kvm/vmx.c
@@ -160,6 +160,7 @@ static void __vcpu_clear(void *arg)
 		vmcs_clear(vcpu->vmcs);
 	if (per_cpu(current_vmcs, cpu) == vcpu->vmcs)
 		per_cpu(current_vmcs, cpu) = NULL;
+	rdtscll(vcpu->host_tsc);
 }
 
 static void vcpu_clear(struct kvm_vcpu *vcpu)
@@ -376,11 +377,16 @@ static void vmx_vcpu_load(struct kvm_vcpu *vcpu)
 {
 	u64 phys_addr = __pa(vcpu->vmcs);
 	int cpu;
+	u64 tsc_this, delta;
 
 	cpu = get_cpu();
 
-	if (vcpu->cpu != cpu)
+	if (vcpu->cpu != cpu) {
 		vcpu_clear(vcpu);
+		rdtscll(tsc_this);
+		delta = vcpu->host_tsc - tsc_this;
+		vmcs_write64(TSC_OFFSET, vmcs_read64(TSC_OFFSET) + delta);
+	}
 
 	if (per_cpu(current_vmcs, cpu) != vcpu->vmcs) {
 		u8 error;

[-- Attachment #3: Type: text/plain, Size: 286 bytes --]

-------------------------------------------------------------------------
This SF.net email is sponsored by DB2 Express
Download DB2 Express C - the FREE version of DB2 express and take
control of your XML. No limits. Just data. Click to get it now.
http://sourceforge.net/powerbar/db2/

[-- Attachment #4: Type: text/plain, Size: 186 bytes --]

_______________________________________________
kvm-devel mailing list
kvm-devel-5NWGOfrQmneRv+LV9MX5uipxlwaOVQ5f@public.gmane.org
https://lists.sourceforge.net/lists/listinfo/kvm-devel

Re: Centrino Duo not working so hot

[David Brown]

> Please check the attached patch (without taskset).  The patch attempts
> to keep the time stamp counter monotonic, even if the guest migrates
> from core to core.

Yeah the patch works with kvm-28 fine with debian testing as a guest
(2.6.18) I'm updating to sid to see if 2.6.21 as a guest has problems
since my netboot install setup doesn't seem to be working so hot with
kvm.

Thanks,
- David Brown

-------------------------------------------------------------------------
This SF.net email is sponsored by DB2 Express
Download DB2 Express C - the FREE version of DB2 express and take
control of your XML. No limits. Just data. Click to get it now.
http://sourceforge.net/powerbar/db2/

Re: Centrino Duo not working so hot

[David Brown]

> Yeah the patch works with kvm-28 fine with debian testing as a guest
> (2.6.18) I'm updating to sid to see if 2.6.21 as a guest has problems
> since my netboot install setup doesn't seem to be working so hot with
> kvm.

Yeah the upgrade to sid went okay so there must be something else
wrong with my netboot setup.

Thanks for the patch when will it make it into the default kernel?

- David Brown

-------------------------------------------------------------------------
This SF.net email is sponsored by DB2 Express
Download DB2 Express C - the FREE version of DB2 express and take
control of your XML. No limits. Just data. Click to get it now.
http://sourceforge.net/powerbar/db2/

Re: Centrino Duo not working so hot

[David Brown]

> Yeah the upgrade to sid went okay so there must be something else
> wrong with my netboot setup.
>
> Thanks for the patch when will it make it into the default kernel?

Uhm, crap... found some more stuff didn't seem to affect the guest OS
or the host for that matter...

This is from dmesg... first line is from when the tap device hooks
into the bridge for the network the rest is when things start
loading...

br0: port 2(tap1) entering learning state
vmwrite error: reg 2010 value 1f92 (err 8208)
 [<f8d9f3d8>] vmcs_write64+0xc/0x1b [kvm_intel]
 [<f8d9fe14>] vmx_vcpu_load+0x5c/0xfa [kvm_intel]
 [<f8de2265>] kvm_vm_ioctl+0x12b/0x7c4 [kvm]
 [<c0150fbe>] __alloc_pages+0x52/0x294
 [<c01c97ba>] prio_tree_insert+0x1d/0x1ef
 [<c0155272>] vma_prio_tree_insert+0x17/0x2a
 [<c015aa5f>] vma_link+0xab/0xc9
 [<c015b75c>] do_mmap_pgoff+0x564/0x6c4
 [<f8de213a>] kvm_vm_ioctl+0x0/0x7c4 [kvm]
 [<c0170ecb>] do_ioctl+0x1f/0x62
 [<c0171152>] vfs_ioctl+0x244/0x256
 [<c01711b0>] sys_ioctl+0x4c/0x64
 [<c0103d8c>] syscall_call+0x7/0xb
 =======================
vmwrite error: reg 2011 value 2010 (err 8209)
 [<f8d9fe14>] vmx_vcpu_load+0x5c/0xfa [kvm_intel]
 [<f8de2265>] kvm_vm_ioctl+0x12b/0x7c4 [kvm]
 [<c0150fbe>] __alloc_pages+0x52/0x294
 [<c01c97ba>] prio_tree_insert+0x1d/0x1ef
 [<c0155272>] vma_prio_tree_insert+0x17/0x2a
 [<c015aa5f>] vma_link+0xab/0xc9
 [<c015b75c>] do_mmap_pgoff+0x564/0x6c4
 [<f8de213a>] kvm_vm_ioctl+0x0/0x7c4 [kvm]
 [<c0170ecb>] do_ioctl+0x1f/0x62
 [<c0171152>] vfs_ioctl+0x244/0x256
 [<c01711b0>] sys_ioctl+0x4c/0x64
 [<c0103d8c>] syscall_call+0x7/0xb
 =======================

These two err 8208 and 8209 repeat for several pages...

Thanks,
- David Brown

-------------------------------------------------------------------------
This SF.net email is sponsored by DB2 Express
Download DB2 Express C - the FREE version of DB2 express and take
control of your XML. No limits. Just data. Click to get it now.
http://sourceforge.net/powerbar/db2/

Re: Centrino Duo not working so hot

[Avi Kivity]

[-- Attachment #1: Type: text/plain, Size: 688 bytes --]

David Brown wrote:
>> Yeah the upgrade to sid went okay so there must be something else
>> wrong with my netboot setup.
>>
>> Thanks for the patch when will it make it into the default kernel?
>
> Uhm, crap... found some more stuff didn't seem to affect the guest OS
> or the host for that matter...
>
> This is from dmesg... first line is from when the tap device hooks
> into the bridge for the network the rest is when things start
> loading...
>
> br0: port 2(tap1) entering learning state
> vmwrite error: reg 2010 value 1f92 (err 8208)

That patch is hopelessly broken; try this instead (revert the old one 
first).

-- 
error compiling committee.c: too many arguments to function


[-- Attachment #2: vmx-tsc-sync.patch --]
[-- Type: text/x-patch, Size: 990 bytes --]

diff --git a/drivers/kvm/kvm_main.c b/drivers/kvm/kvm_main.c
diff --git a/drivers/kvm/vmx.c b/drivers/kvm/vmx.c
index b909b54..08908c0 100644
--- a/drivers/kvm/vmx.c
+++ b/drivers/kvm/vmx.c
@@ -160,6 +160,7 @@ static void __vcpu_clear(void *arg)
 		vmcs_clear(vcpu->vmcs);
 	if (per_cpu(current_vmcs, cpu) == vcpu->vmcs)
 		per_cpu(current_vmcs, cpu) = NULL;
+	rdtscll(vcpu->host_tsc);
 }
 
 static void vcpu_clear(struct kvm_vcpu *vcpu)
@@ -376,6 +377,7 @@ static void vmx_vcpu_load(struct kvm_vcpu *vcpu)
 {
 	u64 phys_addr = __pa(vcpu->vmcs);
 	int cpu;
+	u64 tsc_this, delta;
 
 	cpu = get_cpu();
 
@@ -409,6 +411,13 @@ static void vmx_vcpu_load(struct kvm_vcpu *vcpu)
 
 		rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
 		vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
+
+		/*
+		 * Make sure the time stamp counter is monotonous.
+		 */
+		rdtscll(tsc_this);
+		delta = vcpu->host_tsc - tsc_this;
+		vmcs_write64(TSC_OFFSET, vmcs_read64(TSC_OFFSET) + delta);
 	}
 }
 

[-- Attachment #3: Type: text/plain, Size: 286 bytes --]

-------------------------------------------------------------------------
This SF.net email is sponsored by DB2 Express
Download DB2 Express C - the FREE version of DB2 express and take
control of your XML. No limits. Just data. Click to get it now.
http://sourceforge.net/powerbar/db2/

[-- Attachment #4: Type: text/plain, Size: 186 bytes --]

_______________________________________________
kvm-devel mailing list
kvm-devel-5NWGOfrQmneRv+LV9MX5uipxlwaOVQ5f@public.gmane.org
https://lists.sourceforge.net/lists/listinfo/kvm-devel

Re: Centrino Duo not working so hot

[David Brown]

> That patch is hopelessly broken; try this instead (revert the old one
> first).

Yeah that works fine now, thanks.

- David Brown

-------------------------------------------------------------------------
This SF.net email is sponsored by DB2 Express
Download DB2 Express C - the FREE version of DB2 express and take
control of your XML. No limits. Just data. Click to get it now.
http://sourceforge.net/powerbar/db2/