Re: [PATCH 1/3] kexec: Prevent removal of memory in use by a loaded kexec image

[Baoquan He <bhe@redhat.com>]

On 04/22/20 at 12:05pm, David Hildenbrand wrote:
> On 22.04.20 11:57, Baoquan He wrote:
> > On 04/22/20 at 11:24am, David Hildenbrand wrote:
> >> On 22.04.20 11:17, Baoquan He wrote:
> >>> On 04/21/20 at 03:29pm, David Hildenbrand wrote:
> >>>>>> ACPI SRAT is embeded into efi, need read out the rsdp pointer. If we don't
> >>>>>> pass the efi, it won't get the SRAT table correctly, if I remember
> >>>>>> correctly. Yeah, I remeber kvm guest can get memory hotplugged with
> >>>>>> ACPI only, this won't happen on bare metal though. Need check carefully. 
> >>>>>> I have been using kvm guest with uefi firmwire recently.
> >>>>>
> >>>>> Yeah, I can imagine that bare metal is different. kvm only uses ACPI.
> >>>>>
> >>>>> I'm also asking because of virtio-mem. Memory added via virtio-mem is
> >>>>> not part of any efi tables or whatsoever. So I assume the kexec kernel
> >>>>> will not detect it automatically (good!), instead load the virtio-mem
> >>>>> driver and let it add memory back to the system.
> >>>>>
> >>>>> I should probably play with kexec and virtio-mem once I have some spare
> >>>>> cycles ... to find out what's broken and needs to be addressed :)
> >>>>
> >>>> FWIW, I just gave virtio-mem and kexec/kdump a try.
> >>>>
> >>>> a) kdump seems to work. Memory added by virtio-mem is getting dumped.
> >>>> The kexec kernel only uses memory in the crash region. The virtio-mem
> >>>> driver properly bails out due to is_kdump_kernel().
> >>>
> >>> Right, kdump is not impacted later added memory.
> >>>
> >>>>
> >>>> b) "kexec -s -l" seems to work fine. For now, the kernel does not seem
> >>>> to get placed on virtio-mem memory (pure luck due to the left-to-right
> >>>> search). Memory added by virtio-mem is not getting added to the e820
> >>>> map. Once the virtio-mem driver comes back up in the kexec kernel, the
> >>>> right memory is readded.
> >>>
> >>> kexec_file_load just behaves as you tested. It doesn't collect later
> >>> added memory to e820 because it uses e820_table_kexec directly to pass
> >>> e820 to kexec-ed kernel. However, this e820_table_kexec is only updated
> >>> during boot stage. I tried hot adding DIMM after boot, kexec-ed kernel
> >>> doesn't have it in e820 during bootup, but it's recoginized and added
> >>> when ACPI scanning. I think we should update e820_table_kexec when hot
> >>> add/remove memory, at least for DIMM. Not sure if DLPAR, virtio-mem,
> >>> balloon will need be added into e820_table_kexec too, and if this is
> >>> expected behaviour.
> >>>
> >>> But whatever we do, it won't impact the kexec file_loading, because of
> >>> the searching strategy bottom up. Just adding them into e820_table_kexec
> >>> will make it consistent with cold reboot which get recognizes and get
> >>> them into e820 during bootup.
> >>
> >> Yeah, I think whatever a cold-booted kernel will see is what kexec-ed
> >> kernel should see. Not more, not less.
> >>
> >> Regarding virtio-mem: Not in e820 on cold-boot.
> >> Regarding DIMMs: DIMMs under KVM will never show up in the e820 map
> >> IIRC. I think on real HW it can be different.
> > 
> > Yeah, DIMMs under KVM won't show up in e820 map. While this is not feature
> > of QEMU/KVM, but a defect of it. I ever asked Igor who is developer of
> > QEMU/KVM guest in this area, why we don't make kvm guest recognize
> > hotpluggable DIMM and add it into e820 map, he said he had tried to make
> > it, but this will corrupt guest on HyperV. So he had to revert the
> 
> Yeah, I remember that this had to be reverted due to something breaking.
> But OTOH, it allows us to online coldplugged DIMMs online_movable
> easily, so I'd say it's even a feature (although, does not behave like
> real HW we have).
> 
> I use this extensively when testing memory hot(un)plug via coldplugged
> DIMMs.
> 
> I do wonder if there is real HW, where this is also the case.

None for what I know. Hotplug on real HW includes two parts, the boot
mem being hotpluggable is more flexiable one. It allows people to
replace bad DIMM. And you can see code in boot stage has been adjusted a
lot on this purpose, at that time, people haven't thought about kvm
guest.

> 
> > commit on qemu. So I think we can leave it for now for both real HW and
> > kvm, or update the e820_table_kexec to include added DIMM for both real
> > HW and KVM. I hope one day KVM dev will find a way to conquer the defect
> > on HyperV and make the e820map consistent with bare metal. After all,
> > kvm guest is trying to imitate real HW for the most part.
> > 
> > Anyway, I will think about the e820_table_kexec updating. See if we can
> > do something about it.
> 
> Yeah, for DIMMs on real HW it might definitely make sense. We might be
> able to hook into updates of /sys/firmware/memmap on memory add/remove.