On Thu, Jul 09 2026, David Matlack wrote: > On Thu, May 28, 2026 at 10:41 AM David Matlack <[email protected]> wrote: >> >> This series has 2 small fixes to how FLBs are managed. First is to >> increase the outgoing FLB refcount during liveupdate_flb_get_outgoing() >> so it cannot be freed while the caller is using it, and to align with >> the semantics of liveupdate_flb_get_incoming(). The second is to prevent >> FLB retrieve() from being called multiple times if the first attempt >> fails. >> >> Both of these changes are needed for the correctness of the PCI core >> support for Live Update: >> >> >> https://lore.kernel.org/linux-pci/[email protected]/ > > I'm wondering if we still want this series. After discussions with > Pratyush and Sami (some on-list, some off-list) the v7 of the PCI > series (which I haven't sent yet) does not depend on these changes for > correctness. > > Patch 1: PCI core no longer calls get_outgoing() outside the scope of > file handler callbacks. So there is no need to increment the reference > count. We could still keep this to make get_outgoing behave the same > way as get_incoming for consisteny at the cost of some extra > complexity to manage outgoing FLB refcoutn.
I'm neutral on this one. I'll leave it up to Pasha's preference. > > Patch 2: PCI core panics if retrieve fails (proceeding without the PCI > FLB can lead to memory corruption due to ongoing DMA). So the behavior > of LUO on retrieve() errors does not matter. This patch is useful for sure. Sure, PCI might panic, but other FLB users like HugeTLB won't. And this patch fixes a real problem for those cases. The patch applies cleanly on liveupdate/next. So Mike, would you mind applying patch 2/2 now? We can wait for Pasha to chime in for 1/2. > > Should we drop this series? -- Regards, Pratyush Yadav
