On 05/05/2022 10:52, Vitaly Kuznetsov wrote: > [...] > For Hyper-V, the situation is similar: hv_crash_handler() intitiates > VMbus unload on the crashing CPU only, there's no mechanism to do > 'global' unload so other CPUs will likely not be able to connect Vmbus > devices in kdump kernel but this should not be necessary. > > There's a crash_kexec_post_notifiers mechanism which can be used instead > but it's disabled by default so using machine_ops.crash_shutdown is > better. > Thanks a bunch Vitaly, for the clarification! Just as a heads-up: there's been a panic notifiers refactor proposed [0] in which some notifiers will run before kdump by default, not requiring "crash_kexec_post_notifiers" (which BTW is *unfortunately* hardcoded as 'Y' for hyper-v, since a11589563e96). Cheers, Guilherme [0] https://lore.kernel.org/lkml/20220427224924.592546-1-gpiccoli@xxxxxxxxxx/
