Re: [PATCH v1 1/2] platform/x86/intel: Introduce Intel Elkhart Lake PSE I/O

[Raag Jadav <raag.jadav@intel.com>]

On Wed, Oct 29, 2025 at 10:36:19AM +0200, Andy Shevchenko wrote:
> On Wed, Oct 29, 2025 at 11:50:49AM +0530, Raag Jadav wrote:
> > Intel Elkhart Lake Programmable Service Engine (PSE) includes two PCI
> > devices that expose two different capabilities of GPIO and Timed I/O
> > as a single PCI function through shared MMIO with below layout.
> > 
> > GPIO: 0x0000 - 0x1000
> > TIO:  0x1000 - 0x2000
> > 
> > This driver enumerates the PCI parent device and creates auxiliary child
> > devices for these capabilities. The actual functionalities are provided
> > by their respective auxiliary drivers.
> 
> ...
> 
> > +#include <linux/auxiliary_bus.h>
> > +#include <linux/dev_printk.h>
> > +#include <linux/device/devres.h>
> > +#include <linux/device.h>
> > +#include <linux/err.h>
> > +#include <linux/gfp_types.h>
> > +#include <linux/mod_devicetable.h>
> > +#include <linux/module.h>
> > +#include <linux/pci.h>
> > +#include <linux/sizes.h>
> > +#include <linux/slab.h>
> > +#include <linux/types.h>
> 
> > +#define EHL_PSE_IO_DEV_OFFSET	SZ_4K
> > +#define EHL_PSE_IO_DEV_SIZE	SZ_4K
> 
> Not sure if SZ_4K is a good idea for the _OFFSET, the _SIZE is fine. Also why
> do we need two? If the devices are of the same size, we don't need to have a
> separate offset.

Yes but they're semantically different, atleast as per DEFINE_RES_MEM().
Either way works for me.

...

> > +static int ehl_pse_io_dev_add(struct pci_dev *pci, const char *name, int idx)
> > +{
> > +	struct auxiliary_device *aux_dev;
> > +	struct device *dev = &pci->dev;
> > +	struct ehl_pse_io_dev *io_dev;
> > +	resource_size_t start;
> > +	int ret;
> > +
> > +	io_dev = kzalloc(sizeof(*io_dev), GFP_KERNEL);
> > +	if (!io_dev)
> > +		return -ENOMEM;
> 
> Why devm_kzalloc() can't be used? I don't see if the device lifetime is anyhow
> different to this object. Am I wrong?

Looks like it but I don't know the code well enough to tell if there're
corner cases, so just following the documented rules. Your call.

> > +	start = pci_resource_start(pci, 0);
> > +	io_dev->irq = pci_irq_vector(pci, idx);
> > +	io_dev->mem = DEFINE_RES_MEM(start + (EHL_PSE_IO_DEV_OFFSET * idx), EHL_PSE_IO_DEV_SIZE);
> > +
> > +	aux_dev = &io_dev->aux_dev;
> > +	aux_dev->name = name;
> > +	aux_dev->id = (pci_domain_nr(pci->bus) << 16) | pci_dev_id(pci);
> > +	aux_dev->dev.parent = dev;
> > +	aux_dev->dev.release = ehl_pse_io_dev_release;
> > +
> > +	ret = auxiliary_device_init(aux_dev);
> > +	if (ret)
> > +		goto free_io_dev;
> > +
> > +	ret = __auxiliary_device_add(aux_dev, dev->driver->name);
> 
> Hmm... Is it okay to use double underscored variant? Only a single driver uses
> this so far... Care to elaborate?

The regular variant uses KBUILD_MODNAME which comes with 'intel' prefix
after commit df7f9acd8646, and with that we overshoot the max id string
length for leaf drivers.

> > +	if (ret)
> > +		goto uninit_aux_dev;
> > +
> > +	return 0;
> > +
> > +uninit_aux_dev:
> > +	/* io_dev will be freed with the put_device() and .release sequence */
> 
> Right...
> 
> > +	auxiliary_device_uninit(aux_dev);
> > +free_io_dev:
> > +	kfree(io_dev);
> 
> ...and this is a double free, correct?

Yeah, my sheer incompetence at stealing code :(

Raag