Skip to contentSkip to conversation

disabling "rfkill hotplug" on newer eeepc-laptop models

PATCH Alan Jenkins 3 participants 3 messages Feb 20, 2010 – Feb 22, 2010

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

Hi Matthew,

We've just had the third report of a newer eeepc-laptop model which needs to have the wireless hotplug code disabled. So I expect there are/will be more such models, and many more puzzled users :).

I seem to remember you mentioning two possibilities

1. checking PCI config space, to see whether the device is really absent before we mark it as unplugged.
2. a generic mechanism which would replace the current eeepc-specific code.

I think we should try to implement the former (or equivalent) asap. Is there a generic mechanism yet? If not, I have some eeepc-specific code ready for testing on afflicted models -

drivers/platform/x86/eeepc-laptop.c +0 -0
index 6a47bb7..9a844ca 100644
@@ -578,6 +578,8 @@ static void eeepc_rfkill_hotplug(struct eeepc_laptop *eeepc)
	struct pci_dev *dev;
	struct pci_bus *bus;
	bool blocked = eeepc_wlan_rfkill_blocked(eeepc);

+ bool absent;
+ u32 l;

	if (eeepc->wlan_rfkill)
		rfkill_set_sw_state(eeepc->wlan_rfkill, blocked);
@@ -591,6 +593,22 @@ static void eeepc_rfkill_hotplug(struct eeepc_laptop *eeepc)
			goto out_unlock;
		}

+ if (pci_bus_read_config_dword(bus, 0, PCI_VENDOR_ID, &l)) {
+ pr_err("Unable to read PCI config space?\n");
+ goto out_unlock;
+ }
+ absent = (l == 0xffffffff);
+
+ if (blocked != absent) {
+ pr_warning("BIOS says wireless lan is %s, "
+ "but the pci device is %s\n",
+ blocked ? "blocked" : "unblocked",
+ absent ? "absent" : "present");
+ pr_warning("skipped wireless hotplug as probably "
+ "inappropriate for this model\n");
+ goto out_unlock;
+ }
+

		if (!blocked) {
			dev = pci_get_slot(bus, 0);
			if (dev) {
#1 lore raw
Quoted 3 lines · Alan Jenkins

1. checking PCI config space, to see whether the device is really absent before we mark it as unplugged.
2. a generic mechanism which would replace the current eeepc-specific code.

The generic mechanism will become necessary with .34 anyway, as soon as the PCI runtime power management code lands. It'll be taking the notification handlers for these devices.

I think we effectively need to move some of the acpiphp code and make it generic. These shouldn't be exposed as hotplug slots, but if we receive a rescan notification and the device is no longer there then it's pretty obvious that we need to remove it...

On the other hand, your patch looks fine for the immediate issue.

Signature · 1 line
Matthew Garrett | [email protected]
maintainer
#2 lore raw
Quoted 15 lines · Matthew Garrett
Quoted 3 lines · Alan Jenkins

1. checking PCI config space, to see whether the device is really absent before we mark it as unplugged.
2. a generic mechanism which would replace the current eeepc-specific code.

The generic mechanism will become necessary with .34 anyway, as soon as the PCI runtime power management code lands. It'll be taking the notification handlers for these devices.

I think we effectively need to move some of the acpiphp code and make it generic. These shouldn't be exposed as hotplug slots, but if we receive a rescan notification and the device is no longer there then it's pretty obvious that we need to remove it...

Quoted 5 lines

On the other hand, your patch looks fine for the immediate issue.

--
Matthew Garrett | [email protected]

Alan, could you re-send the patch with a formated changelog ? (and you can add my Acked-By)
Thanks,

Signature · 2 lines
Corentin Chary
http://xf.iksaif.net