Correction and a measurement in place of the guess in the description. The crashkernel reservation on this machine is 2,304 MiB, not 4 GiB: the system reports 127.3 GiB, so the 64G-128G tier applies (2048 MB high + 256 MB low, kexec_crash_size 2415919104).
I tested whether the crashkernel reservation drives the scratch size: three cold boots of the same kernel (7.0.0-38-generic), KHO on by default, with and without crashkernel=. scratch_len stock crashkernel no crashkernel delta lowmem 760 MiB 248 MiB -512 = 2 x 256 MiB (crashkernel low) global 8,510 MiB 4,414 MiB -4,096 = 2 x 2,048 MiB (crashkernel high) per-node 4,338 MiB 4,338 MiB 0 total 13,608 MiB 9,000 MiB -4,608 = 2 x 2,304 MiB Boot log "reserved": 18,849,832K vs 11,771,924K. The difference, 6,912 MiB, is exactly the 2,304 MiB crashkernel plus 4,608 MiB of scratch. So KHO sizes scratch at 2x the early reservations and counts the crashkernel area, which the running kernel never uses. With the default kdump setup, each MiB reserved for kdump costs 3 MiB. Without crashkernel KHO still takes 9.0 GiB here, so disabling the default is still the fix. The crashkernel effect makes it worse on any machine with kdump enabled. -- You received this bug notification because you are a member of Ubuntu Bugs, which is subscribed to Ubuntu. https://bugs.launchpad.net/bugs/2171297 Title: resolute: please disable CONFIG_KEXEC_HANDOVER_ENABLE_DEFAULT (as done for 26.10 in LP: #2169198); KHO scratch makes 13 GiB unusable for GPU / pinned memory To manage notifications about this bug go to: https://bugs.launchpad.net/ubuntu/+source/linux/+bug/2171297/+subscriptions -- ubuntu-bugs mailing list [email protected] https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs
