Re: [PATCH v2 1/1] vfio/pci: Remove the core dependency on driver data

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [Low] pm: dev_pm_ops unconditionally assigned without pm_ptr()
--

commit 3f91e92ded670b6d793f7c9ea8fda1f011d465c3
Author: Simon Song <xinmengs@nvidia.com>
vfio/pci: Remove the core dependency on driver data

This commit introduces VFIO_PCI_CORE_DEFINE_CALLBACKS to generate wrapper
code for each vfio variant driver. It updates variant drivers to store their
private structures in driver_data and use the generated wrappers.

Link: https://lore.kernel.org/all/DLFD2ZDSK9YQ.3A4R66G8UJMD8@kernel.org/

[Severity: Low]
In drivers/vfio/pci/vfio_pci.c, inside the vfio_pci_driver definition
around line 260, the driver pm pointer is assigned directly to
&vfio_pci_pm_ops.

Does this unconditional assignment prevent the compiler from eliminating
the dev_pm_ops structure when CONFIG_PM is disabled? When PM support is
disabled, the dev_pm_ops structures will remain in the compiled binary as
dead code. As mandated by the PM subsystem guidelines, is it possible to
wrap the pointer assignment with pm_ptr() (for example,
.pm = pm_ptr(&vfio_pci_pm_ops)) so the dead structure is safely dropped?

[Severity: Low]
Similarly, in drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c around line
1728, the hisi_acc_vfio_pci_driver assigns the pm field to
&hisi_acc_vf_pm_ops without using the pm_ptr() macro wrapper. Does this
result in the same dead code issue when CONFIG_PM=n?

[Severity: Low]
In drivers/vfio/pci/ism/main.c around line 395, ism_vfio_pci_driver also
unconditionally assigns .pm = &ism_vfio_pm_ops. Should this also be
wrapped with pm_ptr() to ensure the dead code is eliminated?

[Severity: Low]
In drivers/vfio/pci/mlx5/main.c around line 1454, the mlx5vf_pci_driver
assigns .pm = &mlx5vf_pm_ops. Can we use pm_ptr(&mlx5vf_pm_ops) here as
well to avoid leaving an unused structure behind when power management
is disabled?

[Severity: Low]
In drivers/vfio/pci/nvgrace-gpu/main.c around line 1427, the
nvgrace_gpu_vfio_pci_driver directly assigns .pm = &nvgrace_gpu_pm_ops.
Would applying pm_ptr() here ensure consistency with PM guidelines and
allow the structure to be compiled out correctly?

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260930143641.4160785-1-xinmengs@nvidia.com?part=1