On Fri, Nov 13, 2026 at 07:18:55PM +0100, Rafael J. Wysocki (Intel) wrote:
> On Thu, Oct 29, 2026 at 5:31 PM <[email protected]> wrote:
> >
> > From: Frank Li <[email protected]>
> >
> > When CONFIG_PM is not enabled, all devices runtime PM state should be
> > treated as active.
> >
> > __pm_runtime_resume() returns 1 (called by pm_runtime_get()).
> > pm_runtime_get_if_active() is the try-and-get version of pm_runtime_get().
> > stub pm_runtime_active() also returns true. So pm_runtime_get_if_active()
> > should return the same value as the stub pm_runtime_get(), which is 1.
> >
> > Analysis of all callers and their behavior with old (-EINVAL) vs new (1):
> >
> > Pattern A: if (!pm_runtime_get_if_active(dev)) -- no impact
> >   Both -EINVAL and 1 are non-zero, so !ret is false in both cases.
> >   Callers: drivers/accel/ethosu/ethosu_perfmon.c,
> >            drivers/gpu/drm/msm/adreno/a6xx_gmu.c,
> >            drivers/gpu/drm/msm/msm_perfcntr.c,
> >            drivers/gpu/drm/v3d/v3d_mmu.c,
> >            drivers/gpu/drm/v3d/v3d_perfmon.c (x4),
> >            drivers/media/i2c/gc08a3.c gc05a2.c og0ve1b.c ov6211.c,
> >            drivers/media/i2c/ccs/ccs-core.c s5kjn1.c lm3560.c (x2),
> >            drivers/media/i2c/s5k3m5.c thp7312.c imx283.c ov64a40.c,
> >            drivers/media/pci/intel/ipu6/ipu6-buttress.c,
> >            sound/hda/core/device.c
> >
> > Pattern B: if (!ret) after get -- no impact
> >   Both -EINVAL and 1 are non-zero, !ret is false in both cases.
> >   drivers/gpu/drm/amd/pm/amdgpu_pm.c (comment says "allow access"
> >     in kernels without CONFIG_PM), drivers/pci/pci.c
> >
> > Pattern C: ret > 0 / ret <= 0 / ret != 1 checks -- FIXED
> >   drivers/usb/core/offload.c: checks (!= 1), old -EINVAL caused
> >     -EBUSY even with !CONFIG_PM. Fixed: now returns 1, proceeds.
> >   drivers/accel/ivpu/ivpu_job.c: drm_WARN_ON(<= 0), old -EINVAL
> >     spuriously triggered WARN. Fixed: 1 > 0, no warn.
> >   drivers/gpu/drm/i915/intel_runtime_pm.c: checks (<= 0), old
> >     -EINVAL triggered error path. Fixed: 1 > 0, correct path.
> >   drivers/accel/ivpu/ivpu_sysfs.c: (> 0), old skipped freq read.
> >     Fixed: proceeds correctly without CONFIG_PM.
> >   drivers/gpu/drm/xe/xe_pm.c: wraps (> 0), all xe callers
> >     (xe_bo.c, xe_shrinker.c, xe_vsec.c, xe_ggtt.c, xe_guc_ct.c,
> >     xe_pxp.c) now correctly treat !CONFIG_PM as active.
> >   drivers/net/ipa/ipa_smp2p.c: power_on = (> 0), now correctly true.
> >   drivers/iommu/arm/arm-smmu/arm-smmu-qcom.c: (> 0), now correctly
> >     executes stall register update.
> >   drivers/staging/media/ipu7/ipu7-buttress.c: active (<= 0) early
> >     return now skipped, correctly proceeds.
> >
> > Pattern D: explicit -EINVAL check
> >   drivers/hwmon/prom21-xhci.c: checks (> 0) then (== -EINVAL).
> >     Old: took -EINVAL fallback. New: takes (> 0) fast path.
> >     Both succeed when CONFIG_PM is disabled; functionally equivalent.
> >   drivers/hwmon/eic7700-pvt.c: call is guarded by IS_ENABLED(CONFIG_PM),
> >     stub is never reached. No impact.
> >
> > Signed-off-by: Frank Li <[email protected]>
>
> People sent feedback, so I'm wondering if there's going to be a v2 or
> does someone else need to take care of this going forward?

Sorry, I forget this thread. I will post v2 soon.

Frank

Reply via email to