Re: [PATCH v2 01/15] scsi: megaraid_sas: use generic power management

[Bjorn Helgaas]

On Wed, Sep 09, 2020 at 08:50:58PM +0530, Vaibhav Gupta wrote:
> On Wed, Sep 09, 2020 at 08:23:32AM -0500, Bjorn Helgaas wrote:
> > On Wed, Sep 09, 2020 at 03:33:15PM +0530, Vaibhav Gupta wrote:
> > > On Tue, Sep 08, 2020 at 12:32:09PM -0500, Bjorn Helgaas wrote:
> > > > On Mon, Jul 20, 2020 at 07:04:14PM +0530, Vaibhav Gupta wrote:
> > > > > With legacy PM hooks, it was the responsibility of a driver to manage PCI
> > > > > states and also the device's power state. The generic approach is to let
> > > > > the PCI core handle the work.
> > > > > 
> > > > > PCI core passes "struct device*" as an argument to the .suspend() and
> > > > > .resume() callbacks. As the .suspend() work with "struct instance*",
> > > > > extract it from "struct device*" using dev_get_drv_data().
> > > > > 
> > > > > Driver was also using PCI helper functions like pci_save/restore_state(),
> > > > > pci_disable/enable_device(), pci_set_power_state() and pci_enable_wake().
> > > > > They should not be invoked by the driver.
> > > > > 
> > > > > Compile-tested only.
> > > > > 
> > > > > Signed-off-by: Vaibhav Gupta <vaibhavgupta40@xxxxxxxxx>
> > > > > ---
> > > > >  drivers/scsi/megaraid/megaraid_sas_base.c | 61 ++++++-----------------
> > > > >  1 file changed, 16 insertions(+), 45 deletions(-)
> > > > > 
> > > > > diff --git a/drivers/scsi/megaraid/megaraid_sas_base.c b/drivers/scsi/megaraid/megaraid_sas_base.c
> > > > > index 00668335c2af..4a6ee7778977 100644
> > > > > --- a/drivers/scsi/megaraid/megaraid_sas_base.c
> > > > > +++ b/drivers/scsi/megaraid/megaraid_sas_base.c
> > > > > @@ -7539,25 +7539,21 @@ static void megasas_shutdown_controller(struct megasas_instance *instance,
> > > > >  	megasas_return_cmd(instance, cmd);
> > > > >  }
> > > > >  
> > > > > -#ifdef CONFIG_PM
> > > > >  /**
> > > > >   * megasas_suspend -	driver suspend entry point
> > > > > - * @pdev:		PCI device structure
> > > > > - * @state:		PCI power state to suspend routine
> > > > > + * @dev:		Device structure
> > > > >   */
> > > > > -static int
> > > > > -megasas_suspend(struct pci_dev *pdev, pm_message_t state)
> > > > > +static int __maybe_unused
> > > > > +megasas_suspend(struct device *dev)
> > > > >  {
> > > > > -	struct megasas_instance *instance;
> > > > > -
> > > > > -	instance = pci_get_drvdata(pdev);
> > > > > +	struct megasas_instance *instance = dev_get_drvdata(dev);
> > > > >  
> > > > >  	if (!instance)
> > > > >  		return 0;
> > > > >  
> > > > >  	instance->unload = 1;
> > > > >  
> > > > > -	dev_info(&pdev->dev, "%s is called\n", __func__);
> > > > > +	dev_info(dev, "%s is called\n", __func__);
> > > > >  
> > > > >  	/* Shutdown SR-IOV heartbeat timer */
> > > > >  	if (instance->requestorId && !instance->skip_heartbeat_timer_del)
> > > > > @@ -7579,7 +7575,7 @@ megasas_suspend(struct pci_dev *pdev, pm_message_t state)
> > > > >  
> > > > >  	tasklet_kill(&instance->isr_tasklet);
> > > > >  
> > > > > -	pci_set_drvdata(instance->pdev, instance);
> > > > > +	dev_set_drvdata(dev, instance);
> > > > 
> > > > It *might* be correct to replace "instance->pdev" with "dev", but it's
> > > > not obvious and deserves some explanation.  It's true that you can
> > > > replace &pdev->dev with dev, but I don't know anything about
> > > > instance->dev.
> > 
> > Sorry, I meant "instance->pdev" here.
> > 
> > > > I don't think this change is actually necessary, is it?
> > > > "instance->pdev" is still a pci_dev pointer, so pci_set_drvdata()
> > > > should work fine. ...
> > > > 
> > > There is no instance->dev . The 'dev' passed dev_set_drvdata() is
> > > same &pdev->dev. 
> > 
> > Yes, it's true that "dev" here is the same as the "&pdev->dev" we had
> > previously.  But we passed "instance->pdev" (not "pdev") to
> > pci_set_drvdata().  So the question is whether instance->pdev->dev ==
> > dev.
> > 
> > They *might* be the same, but I don't think it's obvious.
> > 
> Yes, they are same.
> driver/pci/pci-driver.c :
> 'dev' is passed as parameter to both pci_device_probe() and pci_pm_suspend()
> From 'dev', pci_device_probe() extracts "struct pci_dev*" and passes it to the
> probe callback of this driver.
> 
> In the proble function - megasas_probe_one()
> :7347	instance->pdev = pdev;
> :7386	pci_set_drvdata(pdev, instance);
> 
> The proble function is using "struct pci_dev*" variable "pdev" provided by core
> and same we replaced &pdev->dev with "struct device *dev".
> 
> So the instance->pdev->dev and 'dev' can only differ if 'dev' passed to
> pci_device_probe() and pci_pm_suspend() are different.

OK.  I think this requires too much analysis for this particular
patch, which is really about deprecating pci_driver.suspend and
.resume.  We want patches to be easily verifiable with a minimum of
extra research.

I would completely support a separate patch to clean this up, though.
The *ideal* thing would be to get rid of the set_drvdata() completely
from the suspend routine.  Doing it in the probe routine should be
enough.

> > > The dev pointer used here, points to same value.
> > > 
> > > pci_get_drvdata() and pci_set_drvdata() invoke dev_get_drvdata() and
> > > dev_set_drvdata() respectively. And they do nothing else. Seems like
> > > additional unnecessary function calls and operations.