On 2026-03-25 07:12 PM, Yi Liu wrote: > > > On 3/25/26 02:00, David Matlack wrote: > > On 2026-03-24 09:07 PM, Yi Liu wrote: > > > On 3/24/26 07:57, David Matlack wrote: > > > > Require that Live Update preserved devices are in singleton iommu_groups > > > > during preservation (outgoing kernel) and retrieval (incoming kernel). > > > > > > > > PCI devices preserved across Live Update will be allowed to perform > > > > memory transactions throughout the Live Update. Thus IOMMU groups for > > > > preserved devices must remain fixed. Since all current use cases for > > > > Live Update are for PCI devices in singleton iommu_groups, require that > > > > as a starting point. This avoids the complexity of needing to enforce > > > > arbitrary iommu_group topologies while still allowing all current use > > > > cases. > > > > > > > > Suggested-by: Jason Gunthorpe <[email protected]> > > > > Signed-off-by: David Matlack <[email protected]> > > > > --- > > > > drivers/pci/liveupdate.c | 34 +++++++++++++++++++++++++++++++++- > > > > 1 file changed, 33 insertions(+), 1 deletion(-) > > > > > > > > diff --git a/drivers/pci/liveupdate.c b/drivers/pci/liveupdate.c > > > > index bec7b3500057..a3dbe06650ff 100644 > > > > --- a/drivers/pci/liveupdate.c > > > > +++ b/drivers/pci/liveupdate.c > > > > @@ -75,6 +75,8 @@ > > > > * > > > > * * The device must not be a Physical Function (PF). > > > > * > > > > + * * The device must be the only device in its IOMMU group. > > > > + * > > > > * Preservation Behavior > > > > * ===================== > > > > * > > > > @@ -105,6 +107,7 @@ > > > > #include <linux/bsearch.h> > > > > #include <linux/io.h> > > > > +#include <linux/iommu.h> > > > > #include <linux/kexec_handover.h> > > > > #include <linux/kho/abi/pci.h> > > > > #include <linux/liveupdate.h> > > > > @@ -222,6 +225,31 @@ static void pci_ser_delete(struct pci_ser *ser, > > > > struct pci_dev *dev) > > > > ser->nr_devices--; > > > > } > > > > +static int count_devices(struct device *dev, void *__nr_devices) > > > > +{ > > > > + (*(int *)__nr_devices)++; > > > > + return 0; > > > > +} > > > > + > > > > > > there was a related discussion on the singleton group check. have you > > > considered the device_group_immutable_singleton() in below link? > > > > > > https://lore.kernel.org/linux-iommu/[email protected]/ > > > > Thanks for the link. > > > > Based on the discussion in the follow-up threads, I think the only check > > in that function that is needed on top of what is in this patch to > > ensure group immutability is this one: > > > > /* > > * The device could be considered to be fully isolated if > > * all devices on the path from the device to the host-PCI > > * bridge are protected from peer-to-peer DMA by ACS. > > */ > > if (!pci_acs_path_enabled(pdev, NULL, REQ_ACS_FLAGS)) > > return false; > > > > However, this would restrict Live Update support to only device > > topologies that have these flags enabled. I am not yet sure if this > > would be overly restrictive for the scenarios we care about supporting. > > yes. It's a bit different from that thread in which not only require > singleton group but also need to be immutable. > > > An alternative way to ensure immutability would be to block adding > > devices at probe time. i.e. Fail pci_device_group() if the device being > > added has liveupdate_incoming=True, or if the group already contains a > > device with liveupdate_{incoming,outgoing}=True. We would still need the > > check in pci_liveupdate_preserve() to pretect against setting > > liveupdate_outgoing=True on a device in a multi-device group. > > this looks good to me. But you'll disallow hotplug-in during liveupdate. > not sure about if any decision w.r.t. hotplug. is it acceptable?
Anyone doing hotplug during the middle of a Live Update is asking for trouble IMO. And it would only prevent a hot-plugged device from coming up if it were to be added to the iommu_group as an existing preserved device. I think that is reasonable. > BTW. A question not specific to this patch. If failure happens after > executing kexec, is there any chance to fallback to the prior kernel? There are many failure paths during the reboot() syscall that can return back to userspace, and then userspace can figure out how to bring the system (e.g. VMs) back online on the current kernel. But otherwise, kexec is currently a one way door. Once you kexec, into the new kernel, you would have to do another Live Update to get back into the previous kernel.
