Hi Damjan, We observed a failure when creating avf interfaces on two types of Arm CPUs, both are SMP system, only one numa-id.
Function vlib_pci_get_device_info() reads '/sys/bus/pci/devices/<devices id>/numa_node' to check which numa_node a NIC device resides in. But on SMP system, -1 is returned as below example, and then later VPP uses -1 to access some arrays which causes memory out-of-bound issue. It seems that -1 is returned where the kernel doesn't have NUMA node information, and the kernel ABI documents -1 as a valid return value here: https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/Documentation/ABI/testing/sysfs-bus-pci#n285 Is it ok to just set di->numa_node to 0, if '/sys/bus/pci/devices/<devices id>/numa_node' returns -1 and there is only one numa node by checking '/sys/devices/system/node/online'? if (-1 == di->numa_node) { if ((err = clib_sysfs_read ("/sys/devices/system/node/online", "%U", unformat_bitmap_list, &bmp))) clib_error_free (err); if (clib_bitmap_count_set_bits (bmp) == 1) di->numa_node = 0; } $ cat /sys/bus/pci/devices/0000:03:00.0/numa_node -1 $ cat /sys/devices/system/node/online 0 $ lscpu Architecture: aarch64 Byte Order: Little Endian CPU(s): 4 On-line CPU(s) list: 0-3 Thread(s) per core: 2 Core(s) per socket: 2 Socket(s): 1 NUMA node(s): 1 Vendor ID: ARM Model: 0 Stepping: r1p0 BogoMIPS: 100.00 L1d cache: 64K L1i cache: 64K L2 cache: 1024K L3 cache: 1024K L4 cache: 8192K NUMA node0 CPU(s): 0-3 Flags: fp asimd evtstrm aes pmull sha1 sha2 crc32 atomics fphp asimdhp cpuid asimdrdm lrcpc dcpop asimddp Thanks.
-=-=-=-=-=-=-=-=-=-=-=- Links: You receive all messages sent to this group. View/Reply Online (#15739): https://lists.fd.io/g/vpp-dev/message/15739 Mute This Topic: https://lists.fd.io/mt/71898582/21656 Group Owner: vpp-dev+ow...@lists.fd.io Unsubscribe: https://lists.fd.io/g/vpp-dev/unsub [arch...@mail-archive.com] -=-=-=-=-=-=-=-=-=-=-=-
