2005-02-10 20:13:27

by Simon White

[permalink] [raw]
Subject: Detecting kernel shutdown in a kernel driver

Hi,

I've been writing a device driver for a piece of hardware that we recently found the pci bridge has an issue on software reset (kernel 2.6.8.1, hardware reset is fine). The bridge appears to corrupt the subvendor/device ids on next boot. We have found a software work around in that I can write to the bridge on module exit and it will always detect correctly next boot (through module_exit when rmmod'd).

However on shutting down a machine with the module loaded it never works next time, so is module_exit actually called?

Secondly I searched through some code and on google to determine if I could detect a shutdown notification in the kernel. I thougt I'd found something using:

static struct pci_driver hsid_driver =
{
.name = HSID_NAME,
.id_table = id_table,
.probe = hsid_probe,
.driver =
{
.shutdown = hsid_shutdown,
},
};

However that also appears not to work. I wondered what the correct solution was for detecting system shutdown in the kernel even if the application using the device has locked up on un-interruptible sleep, so I may try to clean the hardware up a little.

Thankyou for any assistance,
Simon

Please CC me.

--
___________________________________________________________
Sign-up for Ads Free at Mail.com
http://promo.mail.com/adsfreejump.htm


2005-02-10 21:48:49

by Stephen Hemminger

[permalink] [raw]
Subject: Re: Detecting kernel shutdown in a kernel driver

On Thu, 10 Feb 2005 15:05:37 -0500
"Simon White" <[email protected]> wrote:

> Hi,
>
> I've been writing a device driver for a piece of hardware that we recently found the pci bridge has an issue on software reset (kernel 2.6.8.1, hardware reset is fine). The bridge appears to corrupt the subvendor/device ids on next boot. We have found a software work around in that I can write to the bridge on module exit and it will always detect correctly next boot (through module_exit when rmmod'd).
>
> However on shutting down a machine with the module loaded it never works next time, so is module_exit actually called?

(Line wrap your mail please)

> Secondly I searched through some code and on google to determine if I could detect a shutdown notification in the kernel. I thougt I'd found something using:
>
> static struct pci_driver hsid_driver =
> {
> .name = HSID_NAME,
> .id_table = id_table,
> .probe = hsid_probe,
> .driver =
> {
> .shutdown = hsid_shutdown,
> },
> };
>
> However that also appears not to work. I wondered what the correct solution was for detecting system shutdown in the kernel even if the application using the device has locked up on un-interruptible sleep, so I may try to clean the hardware up a little.
>
> Thankyou for any assistance,
> Simon

How about the following, you probably still need pci_hook to handle PCI hot plug,
but you hardware probably doesn't do bus hot plug anyway.

------------

static int hsid_notify(struct notifier_block *this, unsigned long code, void *unused)
{
if (code==SYS_DOWN || code==SYS_HALT) {
bang_the_bridge();
}

return NOTIFY_DONE;
}


static struct notifier_block hsid_notifier = {
.notifier_call = hsid_notify,
};


hsid_module_init()
...
register_reboot_notifier(&hsid_notifier);


hsid_module_exit()
...
unregister_reboot_notifier(&hsid_notifier);

2005-02-11 15:14:51

by Andi Kleen

[permalink] [raw]
Subject: Re: Detecting kernel shutdown in a kernel driver

"Simon White" <[email protected]> writes:

> Hi,
>
> I've been writing a device driver for a piece of hardware that we recently found the pci bridge has an issue on software reset (kernel 2.6.8.1, hardware reset is fine). The bridge appears to corrupt the subvendor/device ids on next boot. We have found a software work around in that I can write to the bridge on module exit and it will always detect correctly next boot (through module_exit when rmmod'd).
>
> However on shutting down a machine with the module loaded it never works next time, so is module_exit actually called?

It's only called when the module is explicitely unloaded.

You can use register_reboot_notifier() though.

-Andi