Hot Dock/Undock support

This series of patches is against the -mm kernel, and will enable docking station support. It is an early patch, but still pretty functional, so I think it's worthwhile to include at this point.

Supported Features:
* Hot Dock/Undock via hardware control
* Enumeration of PCI Devices on Dock Station (Hot Add/Remove) via pci

Not Supported Yet (but will be with laptops with sane dsdts):
* _EJD, _EDL support for devices that aren't enumerable
* hot add of devices other than PCI devices (such as the serial/lpt etc).
* More thorough testing needs to be done for everything, but especially video, as I've not even begun to worry about that.

Please comment on these patches, and test if you have a docking station available. When you find problems, if you would like me to debug them, please load the acpiphp driver with the debug param (modprobe acpiphp debug=1),

This series of patches is against the -mm kernel, and will enable docking station support. It is an early patch, but still pretty functional, so I think it's worthwhile to include at this point. For some laptops, it's necessary to use the pci=assign-busses kernel parameter, because some _DCK methods will attempt to assign bus numbers to the dock bridge (incorrectly).

Hi!

Quoted 6 lines

This series of patches is against the -mm kernel, and will enable docking station support. It is an early patch, but still pretty functional, so I think it's worthwhile to include at this point. For some laptops, it's necessary to use the pci=assign-busses kernel parameter, because some _DCK methods will attempt to assign bus numbers to the dock bridge (incorrectly).

On thinkpad X32: I selected

CONFIG_HOTPLUG_PCI=y
# CONFIG_HOTPLUG_PCI_FAKE is not set
# CONFIG_HOTPLUG_PCI_COMPAQ is not set
# CONFIG_HOTPLUG_PCI_IBM is not set
CONFIG_HOTPLUG_PCI_ACPI=y
# CONFIG_HOTPLUG_PCI_ACPI_IBM is not set
# CONFIG_HOTPLUG_PCI_CPCI is not set
# CONFIG_HOTPLUG_PCI_SHPC is not set
CONFIG_ACPI_IBM=y

Recompiled, rebooted with machine out of dock. /sys/bus/pci/slots/ is empty. I then inserted machine into dock, and locked it:

root@amd:/sys/bus/pci/slots# echo dock > /proc/acpi/ibm/dock
root@amd:/sys/bus/pci/slots# ls
root@amd:/sys/bus/pci/slots#

...still empty. What am I doing wrong?
Pavel

Hi!

There's a bug in error handling:

int __init acpiphp_glue_init(void)
{
int num = 0;

acpi_walk_namespace(ACPI_TYPE_DEVICE, ACPI_ROOT_OBJECT,
ACPI_UINT32_MAX, find_root_bridges, &num,
NULL);

        if (num <= 0)
                return -1;
        else
                acpi_pci_register_driver(&acpi_pci_hp_driver);

        return 0;
}

You register driver here, but if acpiphp_get_num_slots() returns 0 later, you return -ENODEV, aborting load of driver, but without undergistering that driver. It oopses after a while. Same problem if init_slots() returns error.

Pavel

Quoted 8 lines

Quoted 6 lines

This series of patches is against the -mm kernel, and will enable docking station support. It is an early patch, but still pretty functional, so I think it's worthwhile to include at this point. For some laptops, it's necessary to use the pci=assign-busses kernel parameter, because some _DCK methods will attempt to assign bus numbers to the dock bridge (incorrectly).

On thinkpad X32: I selected

Quoted 7 lines

Recompiled, rebooted with machine out of dock. /sys/bus/pci/slots/ is empty. I then inserted machine into dock, and locked it:

root@amd:/sys/bus/pci/slots# echo dock > /proc/acpi/ibm/dock
root@amd:/sys/bus/pci/slots# ls
root@amd:/sys/bus/pci/slots#

Quoted 1 line

...still empty. What am I doing wrong?

You may not use the ibm_acpi driver at the same time as the acpiphp driver for docking. This is because the ibm_acpi driver also tries to handle the dock event notification, and doesn't actually do any hotplugging of the devices. So, you want to set that config option to N. What you are doing above is actually writing to the ibm_acpi driver.

I didn't provide a way to do undocking via software. I just use the button on the dock station. You should however, see something in /sys/bus/pci/slots - can you scan your dmesg to make sure that the acpiphp driver is running? You might run it as a module and enable debugging:

modprobe acpiphp debug=1

This way we can see if it finds your dock slot and registers the notify handler.

Hi!

Quoted 8 lines

Quoted 8 lines

Quoted 6 lines

This series of patches is against the -mm kernel, and will enable docking station support. It is an early patch, but still pretty functional, so I think it's worthwhile to include at this point. For some laptops, it's necessary to use the pci=assign-busses kernel parameter, because some _DCK methods will attempt to assign bus numbers to the dock bridge (incorrectly).

On thinkpad X32: I selected

...

Quoted 7 lines

Quoted 7 lines

Recompiled, rebooted with machine out of dock. /sys/bus/pci/slots/ is empty. I then inserted machine into dock, and locked it:

root@amd:/sys/bus/pci/slots# echo dock > /proc/acpi/ibm/dock
root@amd:/sys/bus/pci/slots# ls
root@amd:/sys/bus/pci/slots#

Quoted 8 lines

Quoted 1 line

...still empty. What am I doing wrong?

You may not use the ibm_acpi driver at the same time as the acpiphp driver for docking. This is because the ibm_acpi driver also tries to handle the dock event notification, and doesn't actually do any hotplugging of the devices. So, you want to set that config option to N. What you are doing above is actually writing to the ibm_acpi driver.

Done.

Quoted 10 lines

I didn't provide a way to do undocking via software. I just use the button on the dock station. You should however, see something in /sys/bus/pci/slots - can you scan your dmesg to make sure that the acpiphp driver is running? You might run it as a module and enable debugging:

modprobe acpiphp debug=1

This way we can see if it finds your dock slot and registers the notify handler.

Result is:

root@amd:/data/l/linux-mm/drivers/pci/hotplug# insmod acpiphp.ko debug=1
insmod: error inserting 'acpiphp.ko': -1 No such device
root@amd:/data/l/linux-mm/drivers/pci/hotplug# dmesg | tail -3
Failure of coda_cnode_make for root: error -19
acpiphp: ACPI Hot Plug PCI Controller Driver version: 0.5
acpiphp_glue: Total 0 slots
root@amd:/data/l/linux-mm/drivers/pci/hotplug#

Should I have CONFIG_HOTPLUG_PCI_ACPI_IBM set?

Pavel

Hum, I don't think so (but maybe someone else knows for sure), I thought that driver was specifically for a certain kind of IBM server, not an IBM laptop. It looks like from this output that the acpiphp is not recognizing any hotplug capable devices on your laptop. I believe that this is defined by acpiphp as a slot which is "ejectable", meaning contains an ACPI method called _EJ0.

Quoted 6 lines · Kristen Accardi

Hum, I don't think so (but maybe someone else knows for sure), I thought that driver was specifically for a certain kind of IBM server, not an IBM laptop. It looks like from this output that the acpiphp is not recognizing any hotplug capable devices on your laptop. I believe that this is defined by acpiphp as a slot which is "ejectable", meaning contains an ACPI method called _EJ0.

To the best of my knowledge, the X-series docking station doesn't contain any PCI devices. It's an extension of the IDE bus plus some broken out serial, parallel, USB and so on. I'd expect driver support for it to just require supporting the dock object and calling its eject routine when the user hits the eject button.

Hum, I don't think so (but maybe someone else knows for sure), I thought that driver was specifically for a certain kind of IBM server, not an IBM laptop. It looks like from this output that the acpiphp is not recognizing any hotplug capable devices on your laptop. I believe that this is defined by acpiphp as a slot which is "ejectable", meaning

contains an ACPI method called _EJ0. I think we should take a look at your dsdt to make sure that it seems reasonable, and also perhaps you could send the output of lspci -vv -x with the laptop booted in the dock just to see what kind of dock bridge you have and make sure everything seems like it should work. Please send the disassembled output of your dsdt - if you've never done it before, instructions for doing this can be found here:
http://acpi.sourceforge.net/dsdt/index.php

Device GDCK looks like dock to my untrained eye. Unfortunately its type is IBM0079... ...

Ahha, and ibm_acpi.c agrees with me.

IBM_HANDLE(dock, root, "\\_SB.GDCK", /* X30, X31, X40 */

        Scope (\_SB)
        {
            Device (GDCK)
            {
                Name (_HID, EisaId ("IBM0079"))
                Name (_CID, 0x150CD041)
                Method (_STA, 0, NotSerialized)
                {
...
                Method (_DCK, 1, NotSerialized)
                {
                    Store (0x00, Local0)
                    If (LEqual (GGID (), 0x03))
                    {
                        Store (\_SB.PCI0.LPC.EC.SDCK (Arg0), Local0)
                    }

                    If (LEqual (GGID (), 0x00))
                    {
                        Store (\_SB.PCI0.PCI1.DOCK.DDCK (Arg0), Local0)
                    }

                    Return (Local0)
                }

                Method (_EJ0, 1, NotSerialized)
                {
                    If (LEqual (GGID (), 0x03))
                    {
                        \_SB.PCI0.LPC.EC.SEJ0 (Arg0)
                    }

                    If (LEqual (GGID (), 0x00))
                    {
                        \_SB.PCI0.PCI1.DOCK.DEJ0 (Arg0)
                    }
                }

....

Hope this helps.
Pavel

On Wed, 2006-01-18 at 23:23 +0100, Pavel Machek wrote:
<snip>

Quoted 44 lines

Device GDCK looks like dock to my untrained eye. Unfortunately its type is IBM0079... ...

Ahha, and ibm_acpi.c agrees with me.

IBM_HANDLE(dock, root, "\\_SB.GDCK", /* X30, X31, X40 */

        Scope (\_SB)
        {
            Device (GDCK)
            {
                Name (_HID, EisaId ("IBM0079"))
                Name (_CID, 0x150CD041)
                Method (_STA, 0, NotSerialized)
                {
...
                Method (_DCK, 1, NotSerialized)
                {
                    Store (0x00, Local0)
                    If (LEqual (GGID (), 0x03))
                    {
                        Store (\_SB.PCI0.LPC.EC.SDCK (Arg0), Local0)
                    }

                    If (LEqual (GGID (), 0x00))
                    {
                        Store (\_SB.PCI0.PCI1.DOCK.DDCK (Arg0), Local0)
                    }

                    Return (Local0)
                }

                Method (_EJ0, 1, NotSerialized)
                {
                    If (LEqual (GGID (), 0x03))
                    {
                        \_SB.PCI0.LPC.EC.SEJ0 (Arg0)
                    }

                    If (LEqual (GGID (), 0x00))
                    {
                        \_SB.PCI0.PCI1.DOCK.DEJ0 (Arg0)
                    }
                }

Quoted 4 lines

....

Hope this helps.
Pavel

so the problem that I see is that this dsdt defines two separate dock devices, one outside the scope of pci, and one within it. The one outside the scope of pci defines the _EJ0 and _DCK methods. So, when acpiphp loads, it scans the pci slots for ejectable slots, finds none (because _EJ0 is defined in the dock device that is outside the scope of pci) and exits. This dsdt is different from the others I've used in that most of them define all methods related to docking under the actual dock bridge (within the scope of pci). perhaps some acpi people can shed some light on the best way to handle this - otherwise I'm sure I can hack something up that will be less than acceptable :).

Quoted 4 lines

Quoted 2 lines

Hope this helps.
Pavel

Quoted 11 lines

so the problem that I see is that this dsdt defines two separate dock devices, one outside the scope of pci, and one within it. The one outside the scope of pci defines the _EJ0 and _DCK methods. So, when acpiphp loads, it scans the pci slots for ejectable slots, finds none (because _EJ0 is defined in the dock device that is outside the scope of pci) and exits. This dsdt is different from the others I've used in that most of them define all methods related to docking under the actual dock bridge (within the scope of pci). perhaps some acpi people can shed some light on the best way to handle this - otherwise I'm sure I can hack something up that will be less than acceptable :).

ACPI has (had?) a braindamage - it drops devices that do not present when initially scanning ACPI namespace. So if you boot undocked - too bad. Driver won't ever see your docking station.

Hi!

Quoted 16 lines

Quoted 15 lines

Quoted 2 lines

Hope this helps.
Pavel

so the problem that I see is that this dsdt defines two separate dock devices, one outside the scope of pci, and one within it. The one outside the scope of pci defines the _EJ0 and _DCK methods. So, when acpiphp loads, it scans the pci slots for ejectable slots, finds none (because _EJ0 is defined in the dock device that is outside the scope of pci) and exits. This dsdt is different from the others I've used in that most of them define all methods related to docking under the actual dock bridge (within the scope of pci). perhaps some acpi people can shed some light on the best way to handle this - otherwise I'm sure I can hack something up that will be less than acceptable :).

Quoted 3 lines

ACPI has (had?) a braindamage - it drops devices that do not present when initially scanning ACPI namespace. So if you boot undocked - too bad. Driver won't ever see your docking station.

I think I booted docked....