On 01.07.20 13:06, David Hildenbrand wrote: > On 01.07.20 13:01, Srikar Dronamraju wrote: >> * David Hildenbrand <david@xxxxxxxxxx> [2020-07-01 12:15:54]: >> >>> On 01.07.20 12:04, Srikar Dronamraju wrote: >>>> * Michal Hocko <mhocko@xxxxxxxxxx> [2020-07-01 10:42:00]: >>>> >>>>> >>>>>> >>>>>> 2. Also existence of dummy node also leads to inconsistent information. The >>>>>> number of online nodes is inconsistent with the information in the >>>>>> device-tree and resource-dump >>>>>> >>>>>> 3. When the dummy node is present, single node non-Numa systems end up showing >>>>>> up as NUMA systems and numa_balancing gets enabled. This will mean we take >>>>>> the hit from the unnecessary numa hinting faults. >>>>> >>>>> I have to say that I dislike the node online/offline state and directly >>>>> exporting that to the userspace. Users should only care whether the node >>>>> has memory/cpus. Numa nodes can be online without any memory. Just >>>>> offline all the present memory blocks but do not physically hot remove >>>>> them and you are in the same situation. If users are confused by an >>>>> output of tools like numactl -H then those could be updated and hide >>>>> nodes without any memory&cpus. >>>>> >>>>> The autonuma problem sounds interesting but again this patch doesn't >>>>> really solve the underlying problem because I strongly suspect that the >>>>> problem is still there when a numa node gets all its memory offline as >>>>> mentioned above. >>>>> >>>>> While I completely agree that making node 0 special is wrong, I have >>>>> still hard time to review this very simply looking patch because all the >>>>> numa initialization is so spread around that this might just blow up >>>>> at unexpected places. IIRC we have discussed testing in the previous >>>>> version and David has provided a way to emulate these configurations >>>>> on x86. Did you manage to use those instruction for additional testing >>>>> on other than ppc architectures? >>>>> >>>> >>>> I have tried all the steps that David mentioned and reported back at >>>> https://lore.kernel.org/lkml/20200511174731.GD1961@xxxxxxxxxxxxxxxxxx/t/#u >>>> >>>> As a summary, David's steps are still not creating a memoryless/cpuless on >>>> x86 VM. >>> >>> Now, that is wrong. You get a memoryless/cpuless node, which is *not >>> online*. Once you hotplug some memory, it will switch online. Once you >>> remove memory, it will switch back offline. >>> >> >> Let me clarify, we are looking for a node 0 which is cpuless/memoryless at >> boot. The code in question tries to handle a cpuless/memoryless node 0 at >> boot. > > I was just correcting your statement, because it was wrong. > > Could be that x86 code maps PXM 1 to node 0 because PXM 1 does neither > have CPUs nor memory. That would imply that we can, in fact, never have > node 0 offline during boot. > Yep, looks like it. [ 0.009726] SRAT: PXM 1 -> APIC 0x00 -> Node 0 [ 0.009727] SRAT: PXM 1 -> APIC 0x01 -> Node 0 [ 0.009727] SRAT: PXM 1 -> APIC 0x02 -> Node 0 [ 0.009728] SRAT: PXM 1 -> APIC 0x03 -> Node 0 [ 0.009731] ACPI: SRAT: Node 0 PXM 1 [mem 0x00000000-0x0009ffff] [ 0.009732] ACPI: SRAT: Node 0 PXM 1 [mem 0x00100000-0xbfffffff] [ 0.009733] ACPI: SRAT: Node 0 PXM 1 [mem 0x100000000-0x13fffffff] -- Thanks, David / dhildenb
