Hi, On 2026-06-24 14:07, Lufei Zheng wrote: > Package: linux > Version: 6.12.86-1 > Severity: important > > Dear Maintainer, > > The riscv64 kernel is built with CONFIG_NR_CPUS=64. On hardware with more > than 64 harts, the excess CPUs are silently lost at boot. > > Hardware: > > Board: SpacemiT V100 (C2201MB1) > Vendor: SpacemiT (mvendorid=0x710) > marchid: 0x8000000058000002 > mimpid: 0x4c4d900 > Arch: riscv64, MMU sv48 > ISA: rv64imafdcvh_zicbom_zicboz_zicntr_zicond_zicsr_zifencei > _zihintntl_zihintpause_zihpm_zimop_zawrs_zfa_zfh_zfhmin > _zca_zcb_zcd_zcmop_zba_zbb_zbc_zbs_zkt_zvbb_zvbc_zve32f > _zve32x_zve64d_zve64f_zve64x_zvfh_zvfhmin_zvkb_zvkg > _zvkned_zvknhb_zvksed_zvksh_zvkt_smaia_smstateen_ssaia > _sscofpmf_sstc_svinval_svnapot_svpbmt > Physical harts: 80 (hart IDs 0-39, 64-87) > > Observed behaviour: > > $ nproc --all > 64 > > $ cat /sys/devices/system/cpu/possible > 0-63 > > 16 harts are silently dropped at boot. The kernel only sees 64 CPUs. > > Boot messages: > > [ 0.000000] NR_CPUS is too small for the number of ACPI tables. > > This message is repeated 16 times. > > Kernel config: > > CONFIG_SMP=y > CONFIG_NR_CPUS=64 > > Expected behaviour: > > All 80 harts should be usable. CONFIG_NR_CPUS should be raised to at least > 256 for riscv64, matching the arm64 default. > > Suggested fix: > > In debian/config/riscv64/config, change: > > CONFIG_NR_CPUS=64 > > to: > > CONFIG_NR_CPUS=256
No, this can't be changed as you say, because the value is not defined here, instead the upstream default is used. Sure we can override it in that file, but it would be way better if we can keep using the upstream default. A MR has been submitted to change the value in the Debian kernel [1] (to 128 instead of 256), but before doing that the default value change should be submitted upstream. Regards Aurelien [1] https://salsa.debian.org/kernel-team/linux/-/merge_requests/1995 -- Aurelien Jarno GPG: 4096R/1DDD8C9B [email protected] http://aurel32.net
