[PATCH v2 02/22] PCI: Add API to track PCI devices preserved across Live Update

[Samiullah Khawaja]

On Mon, Feb 23, 2026 at 03:08:19PM -0800, David Matlack wrote:
>On Mon, Feb 23, 2026 at 2:04 PM Samiullah Khawaja <skhawaja at google.com> wrote:
>> On Thu, Jan 29, 2026 at 09:24:49PM +0000, David Matlack wrote:
>
>> >Drivers can notify the PCI subsystem whenever a device is preserved and
>> >unpreserved with the following APIs:
>> >
>> >  pci_liveupdate_outgoing_preserve(pci_dev)
>> >  pci_liveupdate_outgoing_unpreserve(pci_dev)
>>
>> nit: Preserve and Unpreserve can only be done from outgoing kernel, maybe
>> remove the "outgoing" from the function name.
>
>That's reasonable, I can make that change in v3.

I should have added it earlier, but the same applies to the
pci_liveupdate_incoming_finish() as it can only be done with incoming
kernel. Maybe we can remove incoming from it also for consistency?
>
>> >+static int pci_flb_preserve(struct liveupdate_flb_op_args *args)
>> >+{
>> >+      struct pci_dev *dev = NULL;
>> >+      int max_nr_devices = 0;
>> >+      struct pci_ser *ser;
>> >+      unsigned long size;
>> >+
>> >+      for_each_pci_dev(dev)
>> >+              max_nr_devices++;
>>
>> This will not work for SRIOV as new devices will be registered when
>> sriov is enabled and the max will increase. As we are not handling VFs
>> with this patch series, this can be fine?
>
>That's right. Since this series does not support VFs
>(pci_liveupdate_outgoing_preserve(VF) will return -EINVAL), I dropped
>the logic from v1 that included the maximum possible number of VFs in
>max_nr_devices.
>
>Once we add support for preserving VFs, we can update this logic.
>
>Another way that pci_liveupdate_outgoing_preserve() could fail due to
>running out of room is if hot-plugged devices are preserved. But I
>don't think that's an important problem to solve right now.