On Fri, Feb 02, 2024 at 12:10:58AM +0800, Zhao Liu wrote: > Hi Daniel, > > On Thu, Feb 01, 2024 at 09:21:48AM +0000, Daniel P. Berrangé wrote: > > Date: Thu, 1 Feb 2024 09:21:48 +0000 > > From: "Daniel P. Berrangé" <berrange@xxxxxxxxxx> > > Subject: Re: [PATCH v8 00/21] Introduce smp.modules for x86 in QEMU > > > > On Thu, Feb 01, 2024 at 10:57:32AM +0800, Zhao Liu wrote: > > > Hi Daniel, > > > > > > On Wed, Jan 31, 2024 at 10:28:42AM +0000, Daniel P. Berrangé wrote: > > > > Date: Wed, 31 Jan 2024 10:28:42 +0000 > > > > From: "Daniel P. Berrangé" <berrange@xxxxxxxxxx> > > > > Subject: Re: [PATCH v8 00/21] Introduce smp.modules for x86 in QEMU > > > > > > > > On Wed, Jan 31, 2024 at 06:13:29PM +0800, Zhao Liu wrote: > > > > > From: Zhao Liu <zhao1.liu@xxxxxxxxx> > > > > > > [snip] > > > > > > > > However, after digging deeper into the description and use cases of > > > > > cluster in the device tree [3], I realized that the essential > > > > > difference between clusters and modules is that cluster is an extremely > > > > > abstract concept: > > > > > * Cluster supports nesting though currently QEMU doesn't support > > > > > nested cluster topology. However, modules will not support nesting. > > > > > * Also due to nesting, there is great flexibility in sharing resources > > > > > on clusters, rather than narrowing cluster down to sharing L2 (and > > > > > L3 tags) as the lowest topology level that contains cores. > > > > > * Flexible nesting of cluster allows it to correspond to any level > > > > > between the x86 package and core. > > > > > > > > > > Based on the above considerations, and in order to eliminate the naming > > > > > confusion caused by the mapping between general cluster and x86 module > > > > > in v7, we now formally introduce smp.modules as the new topology level. > > > > > > > > What is the Linux kernel calling this topology level on x86 ? > > > > It will be pretty unfortunate if Linux and QEMU end up with > > > > different names for the same topology level. > > > > > > > > > > Now Intel's engineers in the Linux kernel are starting to use "module" > > > to refer to this layer of topology [4] to avoid confusion, where > > > previously the scheduler developers referred to the share L2 hierarchy > > > collectively as "cluster". > > > > > > Looking at it this way, it makes more sense for QEMU to use the > > > "module" for x86. > > > > I was thinking specificially about what Linux calls this topology when > > exposing it in sysfs and /proc/cpuinfo. AFAICT, it looks like it is > > called 'clusters' in this context, and so this is the terminology that > > applications and users are going to expect. > > The cluster related topology information under "/sys/devices/system/cpu/ > cpu*/topology" indicates the L2 cache topology (CPUID[0x4]), not module > level CPU topology (CPUID[0x1f]). > > So far, kernel hasn't exposed module topology related sysfs. But we will > add new "module" related information in sysfs. The relevant patches are > ready internally, but not posted yet. > > In the future, we will use "module" in sysfs to indicate module level CPU > topology, and "cluster" will be only used to refer to the l2 cache domain > as it is now. So, if they're distinct concepts both relevant to x86 CPUs, then from the QEMU POV, should this patch series be changing the -smp arg to allowing configuration of both 'clusters' and 'modules' for x86 ? An earlier version of this series just supported 'clusters', and this changed to 'modules', but your description of Linux reporting both suggests QEMU would need both. With regards, Daniel -- |: https://berrange.com -o- https://www.flickr.com/photos/dberrange :| |: https://libvirt.org -o- https://fstop138.berrange.com :| |: https://entangle-photo.org -o- https://www.instagram.com/dberrange :|
