On Mon, 19 Mar 2018 15:28:50 +1100 Alexey Kardashevskiy <a...@ozlabs.ru> wrote:
> On 13/3/18 1:26 pm, Alexey Kardashevskiy wrote: > > Hi Igor, > > > > ec78f8114bc4c1 "numa: use possible_cpus for not mapped CPUs check" added a > > warning about "All CPU(s) up to maxcpus should be described in NUMA config, > > ability to start up with partial NUMA mappings is obsoleted and will be > > removed in future" and this is printed when I add a numa node without > > attached CPU like this: > > > > -numa node,nodeid=0,cpus=0,mem=4G > > -numa node,nodeid=1,mem=131072M Warning you see is about CPU(s) not assigned to any node and it has noting to do with cpu-less node. What's the full CLI you are using? > > And the reason for this command line is that I am trying to pass some dodgy > > host RAM (actually belongs to a GPU but directly accessible via a fast > > NVLink, not PCI fabric) which let's say is equally far from all CPUs, at > > least in the host's NUMA config this memory is also not bound to any CPU: -numa node,nodeid=1,mem=131072M - will assign some chunk of initial RAM to cpu-less node. It might serve purpose of cpu-less node simulation and nothing else. You won't be able to pass it GPU memory this way though. It might be possible using -numa node,nodeid=1,memdev=foo but knowing nothing about NVLink, I'm not sure how it's shared with GPU (PCI) and per node memory controllers. > > [aik@aik ~]$ ssh yc02goos numactl -H > > available: 8 nodes (0,8,250-255) > > node 0 cpus: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 > > 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 > > 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 > > node 0 size: 261735 MB > > node 0 free: 258932 MB > > node 8 cpus: 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 > > 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 > > 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 > > 126 127 > > node 8 size: 261739 MB > > node 8 free: 261414 MB > > node 250 cpus: > > node 250 size: 15360 MB > > node 250 free: 15359 MB > > node 251 cpus: > > node 251 size: 0 MB > > node 251 free: 0 MB > > node 252 cpus: > > node 252 size: 15360 MB > > node 252 free: 15359 MB > > node 253 cpus: > > node 253 size: 15360 MB > > node 253 free: 15359 MB > > node 254 cpus: > > node 254 size: 15360 MB > > node 254 free: 15359 MB > > node 255 cpus: > > node 255 size: 15360 MB > > node 255 free: 15359 MB > > node distances: > > node 0 8 250 251 252 253 254 255 > > 0: 10 40 80 80 80 80 80 80 > > 8: 40 10 80 80 80 80 80 80 > > 250: 80 80 10 80 80 80 80 80 > > 251: 80 80 80 10 80 80 80 80 > > 252: 80 80 80 80 10 80 80 80 > > 253: 80 80 80 80 80 10 80 80 > > 254: 80 80 80 80 80 80 10 80 > > 255: 80 80 80 80 80 80 80 10 > > > > > > I am not sure I'll progress far enough to get this working with VFIO but if > > I do, I'd like to keep an ability to have such a partial config in the > > future. What was the reason for this warning in the first place? a CPU(including possible ones) shouldn't hung in nowhere, it should be assigned to some node to describe its relation to memory on nodes. > > Ping? > > >
