NACK: [SRU][J][PATCH 0/1] [UBUNTU 22.04] PCHID per Port Toleration - IBM z17 Enablement (LP: #2119650)

[Massimiliano Pellizzer]

On Thu, 28 Aug 2025 at 11:31, Massimiliano Pellizzer
<massimiliano.pellizzer at canonical.com> wrote:
>
> BugLink: https://bugs.launchpad.net/bugs/2119650
>
> [ Impact ]
>
> Improve how s390 kernel discovers and organizes PCI devices, making it more
> robust, predictable, and compatible with modern hardware and virtualization
> scenarios.
>
> Previously, PCI functions were grouped based on firmware-provided order and
> physical channel IDs (PCHIDs), leading to unstable device numbering and
> incorrect grouping in complex setups.
>
> The new implementation explicitly sorts PCI functions by Requester ID (RID) and
> uses Topology IDs (TIDs) to group multi-function devices, ensuring deterministic
> and future-proof bus/domain creation. It fixes SR-IOV behavior by properly
> grouping Physical Functions (PFs) and Virtual Functions (VFs), allowing PFs
> initially in standby to form shared domains and treating isolated VFs without a
> parent PF as standalone devices.
>
> Parent PF detection has been refactored for consistency, and error handling has
> been improved to prevent leaks and crashes when adding devices dynamically.
>
> [ Fix ]
>
> Backport to Jammy the following commits:
> 52c79e636a58da s390/pci: make better use of zpci_dbg() levels
> 0467cdde8c4320 s390/pci: Sort PCI functions prior to creating virtual busses
> 126034faaac5f3 s390/pci: Use topology ID for multi-function devices
> fd1ae23b495b3a PCI: Prefer 'unsigned int' over bare 'unsigned'
> c3df83e01a96ca PCI: Clean up pci_scan_slot()
> fbed59ed8781d7 PCI: Split out next_ari_fn() from next_fn()
> db360b1ea7faef PCI: Move jailhouse's isolated function handling to
> pci_scan_slot()
> 189c6c33ff421d PCI: Extend isolated function probing to s390
> 960ac362648780 s390/pci: allow zPCI zbus without a function zero
> 45e5f0c017e0d0 s390/pci: clean up left over special treatment for function zero
> 25f39d3dcb48bb s390/pci: Ignore RID for isolated VFs
> 48796104c864cf s390/pci: Fix leak of struct zpci_dev when zpci_add_device()
> fails
> dc287e4c9149ab s390/pci: Fix SR-IOV for PFs initially in standby
> 05793884a1f305 s390/pci: Pull search for parent PF out of
> zpci_iov_setup_virtfn()
> 2844ddbd540fc8 s390/pci: Fix handling of isolated VFs
> 8691abd3afaadd s390/pci: Fix zpci_bus_is_isolated_vf() for non-VFs
> 42420c50c68f3e s390/pci: Fix missing check for zpci_create_device() error return
>
> [ Test Case ]
>
> Testing must be performed on an IBM z17 with Network Express adapters in direct
> mode.
>
> Begin by attaching at least one adapter with two PFs, each mapped to a separate
> port, and confirm that the kernel groups them correctly into distinct domains
> without relying on PCHID ordering.
>
> Next, enable SR-IOV on each PF and verify that child VFs are discovered, grouped
> with the correct PFs, and functional. Test with PFs initially in standby to
> ensure that shared domains are created dynamically. Introduce isolated VFs
> without a visible PF and confirm they enumerate as standalone devices.
>
> Finally, perform hotplug and removal of both PFs and VFs, checking that zpci_dev
> structures are cleaned up correctly without leaks or crashes.
>
> [ Regression Potential ]
>
> The patchset affects s390 PCI subsystem, in particular:
> - device enumeration
> - function grouping
> - SR-IOV handling
> An issue in this code may introduce problems such as incorrect grouping of PFs
> and VFs, unstable bus numbering, or failure to associate VFs with their parent
> PFs.
>
> [ Other Info ]
>
> The patchset has already been tested by IBM using the following PPA:
> - https://launchpad.net/~mpellizzer/+archive/ubuntu/ibm-z
>

-- 
Massimiliano Pellizzer