On Fri, Sep 29, 2023 at 04:54:21PM +0200, Peter Newman wrote: > Hi Tony, > > On Thu, Sep 28, 2023 at 9:14 PM Tony Luck <tony.luck@xxxxxxxxx> wrote: > > diff --git a/Documentation/arch/x86/resctrl.rst b/Documentation/arch/x86/resctrl.rst > > index cb05d90111b4..d6b6a4cfd967 100644 > > --- a/Documentation/arch/x86/resctrl.rst > > +++ b/Documentation/arch/x86/resctrl.rst > > @@ -345,9 +345,15 @@ When control is enabled all CTRL_MON groups will also contain: > > When monitoring is enabled all MON groups will also contain: > > > > "mon_data": > > - This contains a set of files organized by L3 domain and by > > - RDT event. E.g. on a system with two L3 domains there will > > - be subdirectories "mon_L3_00" and "mon_L3_01". Each of these > > + This contains a set of files organized by L3 domain or by NUMA > > + node (depending on whether Sub-NUMA Cluster (SNC) mode is disabled > > + or enabled respectively) and by RDT event. E.g. on a system with > > + SNC mode disabled with two L3 domains there will be subdirectories > > + "mon_L3_00" and "mon_L3_01". The numerical suffix refers to the > > + L3 cache id. With SNC enabled the directory names are the same, > > + but the numerical suffix refers to the node id. > > + Mappings from node ids to CPUs are available in the > > + /sys/devices/system/node/node*/cpulist files. Each of these > > The explanation of mon_data seems overwhelmingly SNC-centric now. > Maybe the SNC section should be responsible for explaining its impact > on the mon_data directory. Mainly by reminding the reader that domain > ids in the mon_data directory are node ids in SNC mode. I cut out all the examples and just note that the numerical suffices are nodes instead of cache instances. This bit of the git diff now reads: - This contains a set of files organized by L3 domain and by - RDT event. E.g. on a system with two L3 domains there will - be subdirectories "mon_L3_00" and "mon_L3_01". Each of these + This contains a set of files organized by L3 domain or by NUMA + node (depending on whether Sub-NUMA Cluster (SNC) mode is disabled + or enabled respectively) and by RDT event. Each of these > > > > directories have one file per event (e.g. "llc_occupancy", > > "mbm_total_bytes", and "mbm_local_bytes"). In a MON group these > > files provide a read out of the current value of the event for > > @@ -452,6 +458,28 @@ and 0xA are not. On a system with a 20-bit mask each bit represents 5% > > of the capacity of the cache. You could partition the cache into four > > equal parts with masks: 0x1f, 0x3e0, 0x7c00, 0xf8000. > > > > +Notes on Sub-NUMA Cluster mode > > +============================== > > +When SNC mode is enabled the "llc_occupancy", "mbm_total_bytes", and > > +"mbm_local_bytes" will only give meaningful results for well behaved NUMA > > +applications. I.e. those that perform the majority of memory accesses > > +to memory on the local NUMA node to the CPU where the task is executing. > > Not being specific about why the results aren't meaningful, this > sounds vague and alarming. Removed the trigger word "meaningful" and re-worded to just explain the increased liklihood that tasks will migrate between nodes, so users must collect data from all nodes. Technically this has always been true on multi-socket systems. But since there is a much higher barrier to task migration between sockets, users may find that simple measurements that used to work now behave differently. New version: +Notes on Sub-NUMA Cluster mode +============================== +When SNC mode is enabled Linux may load balance tasks between Sub-NUMA +nodes much more readily than between regular NUMA nodes since the CPUs +on Sub-NUMA nodes share the same L3 cache and the system may report +the NUMA distance between Sub-NUMA nodes with a lower value than used +for regular NUMA nodes. Users who do not bind tasks to the CPUs of a +specific Sub-NUMA node must read the "llc_occupancy", "mbm_total_bytes", +and "mbm_local_bytes" for all Sub-NUMA nodes where the tasks may execute +to get the full view of traffic for which the tasks were the source. + +The cache allocation feature still provides the same number of +bits in a mask to control allocation into the L3 cache. But each +of those ways has its capacity reduced because the cache is divided +between the SNC nodes. The values reported in the resctrl +"size" files are adjusted accordingly. > > > +Note that Linux may load balance tasks between Sub-NUMA nodes much > > +more readily than between regular NUMA nodes since the CPUs on SNC > > +share the same L3 cache and the system may report the NUMA distance > > +between SNC nodes with a lower value than used for regular NUMA nodes. > > +Tasks that migrate between nodes will have their traffic recorded by the > > +counters in different SNC nodes so a user will need to read mon_data > > +files from each node on which the task executed to get the full > > +view of traffic for which the task was the source. > > + > > + > > +The cache allocation feature still provides the same number of > > +bits in a mask to control allocation into the L3 cache. But each > > +of those ways has its capacity reduced because the cache is divided > > +between the SNC nodes. The values reported in the resctrl > > +"size" files are adjusted accordingly. > > + > > Memory bandwidth Allocation and monitoring > > ========================================== > > > > -- > > 2.41.0 > > > > Reviewed-by: Peter Newman <peternewman@xxxxxxxxxx>
