|
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] Re: [PATCH v13 04/14] xen/percpu: defer freeing per-CPU areas after resume failures
On 16.11.2026 16:15, Mykola Kvach wrote: > Hi Jan, > > Thank you for the review. > > On Tue, Nov 10, 2026 at 1:03 PM Jan Beulich <jbeulich@xxxxxxxx> wrote: >> >> On 27.10.2026 22:05, Mykola Kvach wrote: >>> When a CPU fails to come back during system resume, cpu_up() sends >>> CPU_UP_CANCELED. After trying the other CPUs, enable_nonboot_cpus() >>> sends CPU_RESUME_FAILED for the same CPU. >>> >>> On architectures which do not park offline CPUs, both notifications >>> queue free_percpu_area(). The release metadata, including the rcu_head, >>> is stored in that CPU's per-CPU area. The first callback may already >>> have freed the area when the second notification arrives, or the same >>> rcu_head may be queued twice. The CPU_RESUME_FAILED callbacks also need >>> the area to clean up timers and CPU-pool state. >> >> I think there's another change needed, for the change you presently do >> to not cause a leak: If enable_nonboot_cpus() calls cpu_up() for a CPU >> whose bit isn't set in frozen_cpus, and if that call fails, the CPU >> later wouldn't have the CPU_RESUME_FAILED notification sent for it. >> IOW the freeing would then not happen at all. Question is whether >> simply setting the bit in frozen_cpus would be correct: Notify >> handlers may not expect CPU_RESUME_FAILED for a CPU which wasn't >> online at the point of suspend. > > CPUs outside frozen_cpus are only tried when park_offline_cpus is true. > In that case, the existing !park_offline_cpus check already prevents > freeing their per-CPU areas. > > When park_offline_cpus is false, CPUs outside frozen_cpus are skipped. > A failed cpu_up() leaves the bit set, so CPU_RESUME_FAILED is sent. > > Am I missing a case where this patch would introduce a leak? No, you're right. What I'd like to ask for is an adjustment to the last sentence of the description: "does not take this release path" is ambiguous, because the notifiers are the same everywhere. What's skipped (for both of the notifier codes of interest here) is the call to free_percpu_area(). I'd appreciate if you could disambiguate that. Jan
|
![]() |
Lists.xenproject.org is hosted with RackSpace, monitoring our |