I’ve been working through John Madieu’s Book on Linux Device Driver Development. When typing in the Sample code for the Platform device, I got a Segmentation fault registering the device (insmod).

I started with the example code for registering the platform_driver:

static struct platform_driver mctp_pcc_driver = {
    .probe = mctp_pcc_driver_probe,
    .remove = mctp_pcc_driver_remove,
    
};
static int __init mctp_pcc_mod_init(void)
{
        int rc = 0;
        printk(KERN_INFO "initializing MCTP over PCC\n");
        rc = platform_driver_register(&mctp_pcc_driver);
        return rc;
}

The stack trace in the Kernel Oops showed that there was string compare happening in the platform_driver_register call. Looking at other Kernel modules code, I noticed that it set the .driver { .name } value. This change worked:

static struct platform_driver mctp_pcc_driver = {
    .probe = mctp_pcc_driver_probe,
    .remove = mctp_pcc_driver_remove,
    .driver = {
          .name = "mctp_pcc",
    }
};

Once I made this change, I can insmod the module, and check to see that the system sees the driver:

# find /sys/ -name mctp_pcc
/sys/kernel/btf/mctp_pcc
/sys/kernel/debug/printk/index/mctp_pcc
/sys/bus/platform/drivers/mctp_pcc
/sys/module/mctp_pcc

And rmmod-ing the module removes these /sys entries.

If you go to the Book github site and check the code, you can he he fills out the structure with the owner value as well:

static struct platform_driver mypdrv = {
    .probe      = my_pdrv_probe,
    .remove     = my_pdrv_remove,
    .driver     = {
        .name     = "platform-dummy-char",
        .owner    = THIS_MODULE,
    },
};

One other change I noticed when looking at his sample code is that he does not use an init or exit function for the module, but instead registers the driver with the single macro:

 module_platform_driver(mypdrv);

Which I confirmed also creates and removes the /sys entries upon insmod/rmmod.