[+cc linux-pci]

On Thu, Dec 16, 2021 at 05:30:07PM -0600, Bjorn Helgaas wrote:

[+cc Vaibhav since my main question is about converting to generic PM]

On Thu, May 20, 2021 at 08:55:42PM +0530, Kashyap Desai wrote:

quoted

Signed-off-by: Kashyap Desai <kashyap.desai@broadcom.com>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Tomas Henzl <redacted>
Reviewed-by: Himanshu Madhani <redacted>

Cc: sathya.prakash@broadcom.com
---
 drivers/scsi/mpi3mr/mpi3mr_os.c | 84 +++++++++++++++++++++++++++++++++
 1 file changed, 84 insertions(+)

diff --git a/drivers/scsi/mpi3mr/mpi3mr_os.c b/drivers/scsi/mpi3mr/mpi3mr_os.c
index e63db7c..dd71cdc 100644
--- a/drivers/scsi/mpi3mr/mpi3mr_os.c
+++ b/drivers/scsi/mpi3mr/mpi3mr_os.c

@@ -3389,6 +3389,86 @@ static void mpi3mr_shutdown(struct pci_dev *pdev)
 	mpi3mr_cleanup_ioc(mrioc, 0);
 }
 
+#ifdef CONFIG_PM
+/**
+ * mpi3mr_suspend - PCI power management suspend callback
+ * @pdev: PCI device instance
+ * @state: New power state
+ *
+ * Change the power state to the given value and cleanup the IOC
+ * by issuing MUR and shutdown notification
+ *
+ * Return: 0 always.
+ */
+static int mpi3mr_suspend(struct pci_dev *pdev, pm_message_t state)
+{
+	struct Scsi_Host *shost = pci_get_drvdata(pdev);
+	struct mpi3mr_ioc *mrioc;
+	pci_power_t device_state;
+
+	if (!shost)
+		return 0;

This test of "shost" is unnecessary.  If you set the drvdata before
the .probe() method returns success, you will always get valid drvdata
here.  And you do:

  mpi3mr_probe
    mpi3mr_init_ioc
      mpi3mr_setup_resources
        pci_set_drvdata(pdev, mrioc->shost)

Sure, it's *possible* that a random memory corruption will clear out
the drvdata pointer, but there's no point in testing for that.

quoted

+	mrioc = shost_priv(shost);
+	while (mrioc->reset_in_progress || mrioc->is_driver_loading)
+		ssleep(1);
+	mrioc->stop_drv_processing = 1;
+	mpi3mr_cleanup_fwevt_list(mrioc);
+	scsi_block_requests(shost);
+	mpi3mr_stop_watchdog(mrioc);
+	mpi3mr_cleanup_ioc(mrioc, 1);

This looks possibly wrong.  mpi3mr_cleanup_ioc() takes a "reason",
which looks like it should be MPI3MR_COMPLETE_CLEANUP (0),
MPI3MR_REINIT_FAILURE (1), or MPI3MR_SUSPEND (2).

This should at least use the enum, and it looks like it should use
MPI3MR_SUSPEND instead of passing the MPI3MR_REINIT_FAILURE value.

quoted

+	device_state = pci_choose_state(pdev, state);
+	ioc_info(mrioc, "pdev=0x%p, slot=%s, entering operating state [D%d]\n",
+	    pdev, pci_name(pdev), device_state);
+	pci_save_state(pdev);
+	pci_set_power_state(pdev, device_state);
+	mpi3mr_cleanup_resources(mrioc);

Why is mpi3mr_cleanup_resources() needed here?  It frees IRQs,
iounmaps registers, calls pci_release_selected_regions(), etc.

Very few other drivers do this, and I don't think it's necessary for
suspend.  And if you don't clean it up here, you won't need to
re-initialize it during resume.

I'm asking because I want to convert this from the legacy PCI power
management model to the generic PM model.  The generic model works
like this:

  suspend_devices_and_enter
    dpm_suspend_start(PMSG_SUSPEND)
      pci_pm_suspend                    # PCI bus .suspend() method
        mpi3mr_suspend                  <-- device-specific stuff
    suspend_enter
      dpm_suspend_noirq(PMSG_SUSPEND)
        pci_pm_suspend_noirq            # PCI bus .suspend_noirq() method
          pci_save_state                <-- generic PCI
          pci_prepare_to_sleep          <-- generic PCI
            pci_set_power_state
    ...
    dpm_resume_end(PMSG_RESUME)
      pci_pm_resume                     # PCI bus .resume() method
        pci_restore_standard_config
          pci_set_power_state(PCI_D0)   <-- generic PCI
          pci_restore_state             <-- generic PCI
        mpi3mr_resume                   # dev->driver->pm->resume

Notice that in the generic model the PCI core takes care of
pci_save_state(), pci_set_power_state(), etc., and the device-specific
stuff is done before the generic PCI suspend, and after the generic
PCI resume.

mpi3mr_cleanup_resources() is problematic because the device-specific
stuff should be done *before* the generic PCI part, but you call it
*after* the generic PCI part.  I think it would be *possible* to do it
after by making a mpi3mr_suspend_noirq() method, but there are hardly
any drivers that do that, and I think mpi3mr_cleanup_resources() is
unnecessary here anyway.

quoted

+	return 0;
+}
+
+/**
+ * mpi3mr_resume - PCI power management resume callback
+ * @pdev: PCI device instance
+ *
+ * Restore the power state to D0 and reinitialize the controller
+ * and resume I/O operations to the target devices
+ *
+ * Return: 0 on success, non-zero on failure
+ */
+static int mpi3mr_resume(struct pci_dev *pdev)
+{
+	struct Scsi_Host *shost = pci_get_drvdata(pdev);
+	struct mpi3mr_ioc *mrioc;
+	pci_power_t device_state = pdev->current_state;
+	int r;
+
+	mrioc = shost_priv(shost);
+
+	ioc_info(mrioc, "pdev=0x%p, slot=%s, previous operating state [D%d]\n",
+	    pdev, pci_name(pdev), device_state);
+	pci_set_power_state(pdev, PCI_D0);
+	pci_enable_wake(pdev, PCI_D0, 0);
+	pci_restore_state(pdev);
+	mrioc->pdev = pdev;
+	mrioc->cpu_count = num_online_cpus();
+	r = mpi3mr_setup_resources(mrioc);
+	if (r) {
+		ioc_info(mrioc, "%s: Setup resources failed[%d]\n",
+		    __func__, r);
+		return r;
+	}
+
+	mrioc->stop_drv_processing = 0;
+	mpi3mr_init_ioc(mrioc, 1);
+	scsi_unblock_requests(shost);
+	mpi3mr_start_watchdog(mrioc);
+
+	return 0;
+}
+#endif
+
 static const struct pci_device_id mpi3mr_pci_id_table[] = {
 	{
 		PCI_DEVICE_SUB(PCI_VENDOR_ID_LSI_LOGIC, 0x00A5,

@@ -3404,6 +3484,10 @@ static struct pci_driver mpi3mr_pci_driver = {
 	.probe = mpi3mr_probe,
 	.remove = mpi3mr_remove,
 	.shutdown = mpi3mr_shutdown,
+#ifdef CONFIG_PM
+	.suspend = mpi3mr_suspend,
+	.resume = mpi3mr_resume,
+#endif
 };
 
 static int __init mpi3mr_init(void)

-- 
2.18.1