Re: [PATCH v5 4/8] crash: generic crash hotplug support infrastructure

[Sourabh Jain]

On 15/03/22 19:42, Eric DeVolder wrote:
> On 3/15/22 07:08, Sourabh Jain wrote:
> > Hello Eric,
> >
> > On 03/03/22 21:57, Eric DeVolder wrote:
> > > This patch introduces a generic crash hot plug/unplug infrastructure
> > > for CPU and memory changes. Upon CPU and memory changes, a generic
> > > crash_hotplug_handler() obtains the appropriate lock, does some
> > > important house keeping and then dispatches the hot plug/unplug event
> > > to the architecture specific arch_crash_hotplug_handler(), and when
> > > that handler returns, the lock is released.
> > >
> > > This patch modifies crash_core.c to implement a subsys_initcall()
> > > function that installs handlers for hot plug/unplug events. If CPU
> > > hotplug is enabled, then cpuhp_setup_state() is invoked to register a
> > > handler for CPU changes. Similarly, if memory hotplug is enabled, then
> > > register_memory_notifier() is invoked to install a handler for memory
> > > changes. These handlers in turn invoke the common generic handler
> > > crash_hotplug_handler().
> > >
> > > On the CPU side, cpuhp_setup_state_nocalls() is invoked with parameter
> > > CPUHP_AP_ONLINE_DYN. While this works, when a CPU is being unplugged,
> > > the CPU still shows up in foreach_present_cpu() during the regeneration
> > > of the new CPU list, thus the need to explicitly check and exclude the
> > > soon-to-be offlined CPU in crash_prepare_elf64_headers().
> > >
> > > On the memory side, each un/plugged memory block passes through the
> > > handler. For example, if a 1GiB DIMM is hotplugged, that generate 8
> > > memory events, one for each 128MiB memblock.
> > >
> > > Signed-off-by: Eric DeVolder <eric.devolder@xxxxxxxxxx>
> > > ---
> > >   include/linux/kexec.h |  16 +++++++
> > >   kernel/crash_core.c   | 108 
> > > ++++++++++++++++++++++++++++++++++++++++++
> > >   2 files changed, 124 insertions(+)
> > >
> > > diff --git a/include/linux/kexec.h b/include/linux/kexec.h
> > > index d7b59248441b..b11d75a6b2bc 100644
> > > --- a/include/linux/kexec.h
> > > +++ b/include/linux/kexec.h
> > > @@ -300,6 +300,13 @@ struct kimage {
> > >       /* Information for loading purgatory */
> > >       struct purgatory_info purgatory_info;
> > > +
> > > +#ifdef CONFIG_CRASH_HOTPLUG
> > > +    bool hotplug_event;
> > > +    int offlinecpu;
> > > +    bool elf_index_valid;
> > > +    int elf_index;
> >
> > How about keeping an array to track all kexec segment index need to 
> > be updated in
> > crash hotplug handler.
> >
> > struct hp_segment {
> >     name;
> >     index;
> >     is_valid;
> >   }
> >
> > It will be helpful if architecture need to updated multiple kexec 
> > segments  for a hotplug event.
> >
> > For example, on PowerPC, we might need to update FDT and elfcorehdr 
> > on memory hot plug/unplug.
> >
> > Thanks,
> > Sourabh Jain
>
> Sourabh,
> I'm OK with that. Another idea might be if there are just two, and one 
> of them is elfcorehdr, then perhaps in addition to elf_index and 
> elf_index_valid, maybe we add an arch_index and arch_index_valid? In 
> the case of PPC, the FDT would be stored in arch_index?

Yes it seems like we might not need to keep more than two kexec indexes. 
Since this indexes are arch specific lets push them to struct 
kimage_arch (part of kimage). So for now I will push fdt_index to struct 
kimage_arch.

Thanks,
Sourabh Jain