Hello !

I am looking to get some feedback and evaluation on a performance-related patch for Xen (XCP-ng 8.3 only).
This patch changes the memcpy implementation of Xen to use the "ERMS variant" (aka REP MOVSB) instead of the current REP MOVSQ+B implementation.
This is expected to perform better on the vast majority of Intel CPUs and modern AMD ones (Zen3+), but may perform worse on some older AMD CPUs.

This change may impact the performance of PV drivers (especially network).

You can find more details regarding this proposed change in : https://github.com/xcp-ng-rpms/xen/pull/54
This change may be reworked in the future to take more in account the specificities of each CPUs (e.g check presence of ERMS flag).

Keep in mind that this patched version is experimental and not officially supported.

Installation :

# Download repo file for XCP-ng 8.3
wget https://koji.xcp-ng.org/repos/user/8/8.3/xcpng-users.repo -O /etc/yum.repos.d/xcpng-users.repo

# Installing the patched Xen packages (you should see `.erms` packages)
yum update --enablerepo=xcp-ng-tae1

You can revert the changes by downgrading the Xen package with the ones in the default repos.

yum downgrade --disablerepo=xcp-ng-tae1 "xen-*"

TSnake41 opened this pull request in xcp-ng-rpms/xen

draft Use ERMS variant for memcpy #54

@plaidypus I don't know a lot about NUMA on Xen, but we have a part in the docs regarding that
https://docs.xcp-ng.org/compute/#numa-affinity

And also other documentation on the subject
https://xapi-project.github.io/new-docs/toolstack/features/NUMA/index.html
there was a design session regarding NUMA in latest Xen Summit : https://youtu.be/KoNwEYMlhyU?list=PLQMQQsKgvLnvjRgDnb-5T51e1kGHgs1SO

@tuxen said (https://xcp-ng.org/forum/topic/3652/no-free-virtual-function-found-vgpu-s7150/4?_=1731502751059)

After some digging, could be the case of a GPU firmware being incompatible with UEFI. Do you have any spare server for testing XCP-ng boot in legacy/BIOS with this GPU?

Perhaps it is the issue ?

@ohajek

Nov 13 11:30:21 xen03 kernel: [10188.720655] AMD IOMMUv2 driver by Joerg Roedel jroedel@suse.de
Nov 13 11:30:21 xen03 kernel: [10188.720656] AMD IOMMUv2 functionality not available on this system

This is expected, Dom0 Kernel (Linux) is not supposed to access the IOMMU when it is already used by Xen. To check if AMD-Vi is working, you need to check xl dmesg instead.

I took a quick look at kern_gim_compiled.txt, and it look likes it timed-out somewhere

Oct 23 20:49:32 xen03 kernel: [   80.657394]        gim error:(wait_cmd_complete:2387)  wait_cmd_complete -- time out after 0.003004460 sec
Oct 23 20:49:32 xen03 kernel: [   80.657408]        gim error:(wait_cmd_complete:2390)   Cmd = 0x17, Status = 0x0, cmd_Complete=0

3ms looks like a short timeout for me, but aside that, it looks like a driver(gim) or hardware issue

@abudef Note that even with this update, nested virtualization is still not really supported in XCP-ng 8.3.
It's there, you can enable it at your own risk. It broke due to some change in XAPI (even though Xen hypervisor had "support" for it).
It never actually got removed from Xen hypervisor (it was marked experimental in Xen 4.13 used in XCP-ng 8.2, it is also the case for Xen 4.17), although nothing really changed, it still has the same issues and limitations as said.

The current state of nested virtualization in Xen is quite clumsy and there are future plans to remake it properly from ground without taking shortcuts and have proper tests to back it.

Aside that, after some experiments, it seems that mostly nested EPT is incomplete/buggy, so your L1 hypervisor should not rely on it. You should add hap=0 to nested XCP-ng Xen cmdline. Beware that it will imply a pretty large performance hit, but I had more consistent results with this.
I am quite suprised that Windows works while Linux don't, maybe it is somewhat related to PV drivers ?

PCIe AER needs proper PCIe, which in practice needs Q35 chipset in the guest (or some other guest type/PCI passthrough way).

Q35 support is currently work in progress

I think it is the same MSI-X/PBA issues that may be partially fixed with https://gitlab.com/xen-project/xen/-/commit/b2cd07a0447bfa25e96ae13e190225b61a3670cb

However, with this device, MSI-X vector table and PBA are in a same page (vector table in 46800 and PBA in 46068) though, which is threated a bit differently

If PBA lives on the same page, discard writes and log a message.
Technically, writes outside of PBA could be allowed, but at this moment
the precise location of PBA isn't saved, and also no known device abuses
the spec in this way (at least yet).

But Coral appears to abuse this according to DKMS driver by having more than just MSI-X and PBA on a single page
https://github.com/google/gasket-driver/blob/main/src/apex_driver.c#L103-L140

Hello @hvm,

Can you give the output of xl dmesg in XCP-ng and of dmesg in the guest that has the issues ?
I have the impression that something is going wrong with reserved regions related to the SATA controller.

@olivierlambert

iperf (-P8) for VM to VM on Xeon Gold 6138
Before: 25-35 Gbps
After: 25-39 Gbps

Seems slightly higher at best, but hard to measure a difference as the performance tend to vary a lot between runs.

@vkeven
XCP-ng 8.1 release note says

VSS and quiesced snapshots support is removed, because it never worked correctly and caused more harm than good. Note that Windows guest tools version 9 (the default for recent versions of Windows if you install Citrix drivers) already removed VSS support, even for older versions of CH / XCP-ng

I am not sure if this VSS feature is bound to the PV drivers, or if it also needs hypervisor support. Though it is not recommended to stay on a old version of the guest agent.

Hello !

I am looking to get some feedback and evaluation on a performance-related patch for Xen (XCP-ng 8.3 only).
This patch changes the memcpy implementation of Xen to use the "ERMS variant" (aka REP MOVSB) instead of the current REP MOVSQ+B implementation.
This is expected to perform better on the vast majority of Intel CPUs and modern AMD ones (Zen3+), but may perform worse on some older AMD CPUs.

This change may impact the performance of PV drivers (especially network).

You can find more details regarding this proposed change in : https://github.com/xcp-ng-rpms/xen/pull/54
This change may be reworked in the future to take more in account the specificities of each CPUs (e.g check presence of ERMS flag).

Keep in mind that this patched version is experimental and not officially supported.

Installation :

# Download repo file for XCP-ng 8.3
wget https://koji.xcp-ng.org/repos/user/8/8.3/xcpng-users.repo -O /etc/yum.repos.d/xcpng-users.repo

# Installing the patched Xen packages (you should see `.erms` packages)
yum update --enablerepo=xcp-ng-tae1

You can revert the changes by downgrading the Xen package with the ones in the default repos.

yum downgrade --disablerepo=xcp-ng-tae1 "xen-*"

TSnake41 opened this pull request in xcp-ng-rpms/xen

draft Use ERMS variant for memcpy #54

@nvs said in XCP-NG server crashes/reboots unexpectedly:

Thanks. Unfortunately my machine doesnt have IPMI. So can I just connect a serial cable between this machine and another machine

Yes though you would still need to boot using the "XCP-ng (Serial)" grub entry.
(you can also add some serial console bits adding them to xen cmdline)

@nvs said in XCP-NG server crashes/reboots unexpectedly:

@nvs Machine crashed/restarted itself again this morning. I didn't even have all of the usual VMs running this time. Nothing was logged in kern.log when it crashed again. Before it crashed I checked a few times in the hours before xl dmesg but nothing obvious to me (same log as I posted above). Any suggestions highly welcome as I'm sure how to proceed with troubleshooting this. My next step would be replacing the PSU and see if anything changes, but its a long shot.

Ok so it doesn't seem caused by a driver bug causing corruption somewhere in Xen/Linux.
So something is causing Xen to crash, and it's not very easy to know without using e.g the serial console (so you can get the actual Xen crash message).
You need for that something connected and monitoring the machine's serial console (or using IPMI) and boot XCP-ng in "(serial)" mode.

@nvs you need to reboot, and it should stick accros reboots

Side question, any chance to tail -f xl dmesg to see real time output? That would allow me to see any last messages before it crashes potentially.

xl dmesg -c allow you to clear the Xen console while displaying it

Can you try adding iommu=strict to Xen command-line ?

/opt/xensource/libexec/xen-cmdline --set-xen "dom0-iommu=strict"

And regularly check if there is something showing up in xl dmesg.

@JeffBerntsen said in Status of PVH:

@TeddyAstie Do you know if these can be set in XO or are they only via command line at this point? I'm curious and also curious as to how well supported because I'm looking at a possible migration from XS 7.1.2 to XCP-ng 8.3 where there are several 32-bit PV machines.

It can be only be set from command-line for now. Though Xen Orchestra should be updated to convert to PV-in-PVH instead of between HVM and PV (this one being unsupported and not bootable anymore).

@sborrill ouch that not practical
The command you provide doesn't work for pv-in-pvh.
A equivalent for the PV-shim would be

xe vm-param-add uuid={UUID} param-name=platform pvinpvh-xen-cmdline="pv-shim console=xen pv-linear-pt=true"

We don't have a usable support for PVH guest at this time, though, for booting a PV guest on XCP-ng 8.3, you can enable PV-shim, e.g

xe vm-param-set uuid={UUID} domain-type=pv-in-pvh