Re: [net-next 04/15] ice: Add advanced power mgmt for WoL

[Lukas Wunner <lukas@wunner.de>]

On Thu, Jul 23, 2020 at 04:47:09PM -0700, Tony Nguyen wrote:
> From: Akeem G Abodunrin <akeem.g.abodunrin@intel.com>
> 
> Add callbacks needed to support advanced power management for Wake on LAN.
> Also make ice_pf_state_is_nominal function available for all configurations
> not just CONFIG_PCI_IOV.

The above was applied as commit 769c500dcc1e.

> +static int ice_resume(struct device *dev)
> +{
> +	struct pci_dev *pdev = to_pci_dev(dev);
> +	enum ice_reset_req reset_type;
> +	struct ice_pf *pf;
> +	struct ice_hw *hw;
> +	int ret;
> +
> +	pci_set_power_state(pdev, PCI_D0);
> +	pci_restore_state(pdev);
> +	pci_save_state(pdev);
> +
> +	if (!pci_device_is_present(pdev))
> +		return -ENODEV;
> +
> +	ret = pci_enable_device_mem(pdev);
> +	if (ret) {
> +		dev_err(dev, "Cannot enable device after suspend\n");
> +		return ret;
> +	}

You're calling pci_enable_device_mem() on resume without having called
pci_disable_device() on suspend.  This leads to an imbalance of the
enable_cnt kept internally in the PCI core.

Every time you suspend, the enable_cnt keeps growing.

The user-visible effect is that if you suspend the device at least once
and then unbind the driver, pci_disable_device() isn't called because
the enable_cnt hasn't reached zero (and will never reach it again).

I recommend removing the call to pci_enable_device_mem() in ice_resume():
The call to pci_restore_state() should already be sufficient to set the
Memory Space bit in the Command register again on resume.

I cannot test this for lack of hardware but can provide a patch if you
want me to.

Thanks,

Lukas