Re: [RFC PATCH 01/13] PCI: Prepare removing devres from pci_intx()

[Andy Shevchenko <andriy.shevchenko@xxxxxxxxxxxxxxx>]

On Wed, Oct 09, 2024 at 10:35:07AM +0200, Philipp Stanner wrote:
> pci_intx() is a hybrid function which sometimes performs devres
> operations, depending on whether pcim_enable_device() has been used to
> enable the pci_dev. This sometimes-managed nature of the function is
> problematic. Notably, it causes the function to allocate under some
> circumstances which makes it unusable from interrupt context.
> 
> To, ultimately, remove the hybrid nature from pci_intx(), it is first
> necessary to provide an always-managed and a never-managed version
> of that function. Then, all callers of pci_intx() can be ported to the
> version they need, depending whether they use pci_enable_device() or
> pcim_enable_device().
> 
> An always-managed function exists, namely pcim_intx(), for which
> __pcim_intx(), a never-managed version of pci_intx() had been
> implemented.

> Make __pcim_intx() a public function under the name
> pci_intx_unmanaged(). Make pcim_intx() a public function.

To avoid an additional churn we can make just completely new APIs, namely:
pcim_int_x()
pci_int_x()

You won't need all dirty dances with double underscored function naming and
renaming.


...

> +	pci_read_config_word(pdev, PCI_COMMAND, &pci_command);
> +
> +	if (enable)
> +		new = pci_command & ~PCI_COMMAND_INTX_DISABLE;
> +	else
> +		new = pci_command | PCI_COMMAND_INTX_DISABLE;
> +
> +	if (new != pci_command)

I would use positive conditionals as easy to read (yes, a couple of lines
longer, but also a win is the indentation and avoiding an additional churn in
the future in case we need to add something in this branch.

> +		pci_write_config_word(pdev, PCI_COMMAND, new);

...

Otherwise I'm for the idea in general.

-- 
With Best Regards,
Andy Shevchenko