| Message ID | 20240403105511.558395-1-andriy.shevchenko@linux.intel.com (mailing list archive) |
|---|---|
| State | Rejected, archived |
| Headers | show |
| Series | [v1,1/1] platform/x86/intel: atomisp2: Replace deprecated UNIVERSAL_DEV_PM_OPS() | expand |
Hi Andy, On 4/3/24 12:55 PM, Andy Shevchenko wrote: > The UNIVERSAL_DEV_PM_OPS() macro is deprecated. Replace it. > > Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com> As mentioned in the description of DEFINE_RUNTIME_DEV_PM_OPS() DEFINE_RUNTIME_DEV_PM_OPS() is NOT a 1:1 replacement for UNIVERSAL_DEV_PM_OPS() specifically it uses pm_runtime_force_suspend() / pm_runtime_force_resume() . Specifically pm_runtime_force_suspend() may NOT get set (and in this case will not set) needs_force_resume skipping a resume + suspend cycle after a system suspend, which is a problem if firmware has touched the state of the device during the suspend/resume cycle since the device may now actually be left powered on. It seems there is no direct replacement for UNIVERSAL_DEV_PM_OPS() without a behavior change. Regards, Hans > --- > drivers/platform/x86/intel/atomisp2/pm.c | 5 ++--- > 1 file changed, 2 insertions(+), 3 deletions(-) > > diff --git a/drivers/platform/x86/intel/atomisp2/pm.c b/drivers/platform/x86/intel/atomisp2/pm.c > index 805fc0d8515c..1081b632bd5e 100644 > --- a/drivers/platform/x86/intel/atomisp2/pm.c > +++ b/drivers/platform/x86/intel/atomisp2/pm.c > @@ -118,8 +118,7 @@ static int isp_pci_resume(struct device *dev) > return 0; > } > > -static UNIVERSAL_DEV_PM_OPS(isp_pm_ops, isp_pci_suspend, > - isp_pci_resume, NULL); > +static DEFINE_RUNTIME_DEV_PM_OPS(isp_pm_ops, isp_pci_suspend, isp_pci_resume, NULL); > > static const struct pci_device_id isp_id_table[] = { > { PCI_VDEVICE(INTEL, 0x0f38), }, > @@ -133,7 +132,7 @@ static struct pci_driver isp_pci_driver = { > .id_table = isp_id_table, > .probe = isp_probe, > .remove = isp_remove, > - .driver.pm = &isp_pm_ops, > + .driver.pm = pm_ptr(&isp_pm_ops), > }; > > module_pci_driver(isp_pci_driver);
On Mon, Apr 08, 2024 at 06:20:03PM +0200, Hans de Goede wrote: > On 4/3/24 12:55 PM, Andy Shevchenko wrote: ... > As mentioned in the description of DEFINE_RUNTIME_DEV_PM_OPS() > DEFINE_RUNTIME_DEV_PM_OPS() is NOT a 1:1 replacement for > UNIVERSAL_DEV_PM_OPS() specifically it uses pm_runtime_force_suspend() / > pm_runtime_force_resume() . Right. > Specifically pm_runtime_force_suspend() may NOT get set (and in this case > will not set) needs_force_resume skipping a resume + suspend cycle > after a system suspend, which is a problem if firmware has touched > the state of the device during the suspend/resume cycle since the device > may now actually be left powered on. I see, thanks for explaining me this. So this driver is kinda very special. Still the old question, can we get rid altogether of these atomisp "drivers" in PDx86? > It seems there is no direct replacement for UNIVERSAL_DEV_PM_OPS() > without a behavior change. Correct. ... Btw, have you seen a few cleanup patches against AtomISP v2 by me?
Hi, On 4/9/24 7:15 PM, Andy Shevchenko wrote: > On Mon, Apr 08, 2024 at 06:20:03PM +0200, Hans de Goede wrote: >> On 4/3/24 12:55 PM, Andy Shevchenko wrote: > > ... > >> As mentioned in the description of DEFINE_RUNTIME_DEV_PM_OPS() >> DEFINE_RUNTIME_DEV_PM_OPS() is NOT a 1:1 replacement for >> UNIVERSAL_DEV_PM_OPS() specifically it uses pm_runtime_force_suspend() / >> pm_runtime_force_resume() . > > Right. > >> Specifically pm_runtime_force_suspend() may NOT get set (and in this case >> will not set) needs_force_resume skipping a resume + suspend cycle >> after a system suspend, which is a problem if firmware has touched >> the state of the device during the suspend/resume cycle since the device >> may now actually be left powered on. > > I see, thanks for explaining me this. So this driver is kinda very special. > Still the old question, can we get rid altogether of these atomisp "drivers" > in PDx86? At some time in the future yes. I've recently done some improvements to the staging atomisp driver so that it will run in a pm-only mode when the firmware is missing so that the ISP still gets turned off properly in this case and the driver now supports both BYT + CHT in a single build so in a way it is ready to replace the atomisp2 pm driver. But it is still in staging, so distros are unlikely to enable it and without a atomisp2-pm driver the battery drains much to quickly especially when suspended. So I think we are getting there, but now is not the right moment to drop this driver. >> It seems there is no direct replacement for UNIVERSAL_DEV_PM_OPS() >> without a behavior change. > > Correct. > > ... > > Btw, have you seen a few cleanup patches against AtomISP v2 by me? Yes I have a bit of a backlog, I have just processed them, thank you for those patches. Regards, Hans
diff --git a/drivers/platform/x86/intel/atomisp2/pm.c b/drivers/platform/x86/intel/atomisp2/pm.c index 805fc0d8515c..1081b632bd5e 100644 --- a/drivers/platform/x86/intel/atomisp2/pm.c +++ b/drivers/platform/x86/intel/atomisp2/pm.c @@ -118,8 +118,7 @@ static int isp_pci_resume(struct device *dev) return 0; } -static UNIVERSAL_DEV_PM_OPS(isp_pm_ops, isp_pci_suspend, - isp_pci_resume, NULL); +static DEFINE_RUNTIME_DEV_PM_OPS(isp_pm_ops, isp_pci_suspend, isp_pci_resume, NULL); static const struct pci_device_id isp_id_table[] = { { PCI_VDEVICE(INTEL, 0x0f38), }, @@ -133,7 +132,7 @@ static struct pci_driver isp_pci_driver = { .id_table = isp_id_table, .probe = isp_probe, .remove = isp_remove, - .driver.pm = &isp_pm_ops, + .driver.pm = pm_ptr(&isp_pm_ops), }; module_pci_driver(isp_pci_driver);
The UNIVERSAL_DEV_PM_OPS() macro is deprecated. Replace it. Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com> --- drivers/platform/x86/intel/atomisp2/pm.c | 5 ++--- 1 file changed, 2 insertions(+), 3 deletions(-)