Re: [PATCH 0/4, v3] Physical PCI slot objects

[Kenji Kaneshige <kaneshige.kenji@jp.fujitsu.com>]

Hi Kristen,

> Hello - back to this mail for a moment.  I wanted to talk about acpiphp
> again.  You say that acpiphp has this same problem as Alex's driver wrt
> evalating _SUN on a non existant object.  If we skip registering the
> slot in acpiphp's register_slot function if it returns 0 for _STA, we
> will not detect any hotplug slots, since leaf nodes will return 0 for
> _STA if a device is not present. Should we only skip it if it's parent
> is a p2p bridge that has _EJ0 and the parent's _STA says empty?

I've been not looking at acpiphp code for a long time. So I need refresh
my memory before answering. Sorry.

But I want to say one thing. We can detect all the hotplug slots if ACPI
firmware doesn't provide _STA. We don't need _STA for PCI hotplug slots
and should not provide _STA for PCI hotplug slots if the system is based
on ACPI2.0 or later, because those are _ADR based devices (PCI root bridge,
which is _HID based device, is a exception). For _ADR based devices, we
should check the presence of the device using PCI bus drivers instead of
evaluating _STA. Please see the section of "_EJx (Eject)" in ACPI2.0b or
later spec. It says "For _HID devices, OSPM evaluates the _STA method. For
_ADR devices, OSPM checks with the bus driver for that device.".

BTW, acpiphp_glue.c says _STA is "optionally". I don't know what it means.

Thanks,
Kenji Kaneshige


Kristen Carlson Accardi wrote:
> On Thu, 29 Nov 2007 16:47:14 +0900
> Kenji Kaneshige <kaneshige.kenji@jp.fujitsu.com> wrote:
> 
>>> Hi Gary, Kenji-san, et. al,
>>>
>>> * Gary Hade <garyhade@us.ibm.com>:
>>>> Alex, What I was trying to suggest is a boot-time kernel
>>>> option, not a kernel configuration option.  The basic idea is
>>>> to give the user (with a single binary kernel) the ability to
>>>> include your ACPI-PCI slot driver feature changes only when
>>>> they are really needed.  In addition to reducing the number of
>>>> system/PCI hotplug driver combinations where your changes would
>>>> need to be validated, I believe would also help alleviate other
>>>> worries (e.g. Andi Kleen's memory consumption concern).  I
>>>> believe this goal could also be achieved with the kernel config
>>>> option by making the pci_slot module runtime loadable with the
>>>> PCI hotplug drivers only visiting your new code when the
>>>> pci_slot driver is loaded, although I think this would be more
>>>> difficult to implement.
>>> I have modified my patch series so that the final patch that
>>> introduces my ACPI-PCI slot driver is a full-fledged module, that
>>> has a tristate Kconfig option.
>>>
>> Thank you for your good job.
>>
>> I tested shpchp and pciehp both with and without pci_slot module.
>> There seems no regression from shpchp and pciehp's point of view.
>> (I had a little concern about the hotplug slots' name that vary
>> depending on whether pci_slot functionality is enabled or disabled.
>> But, now that we can build pci_slot driver as a kernel module, I
>> don't think it is a big problem).
>>
>> Only the problems is that I got Call Traces with the following error
>> messages when pci_slot driver was loaded, and one strange slot named
>> '1023' was registered (other slots are fine). This is the same problem
>> I reported before.
>>
>>     sysfs: duplicate filename '1023' can not be created
>>     WARNING: at fs/sysfs/dir.c:424 sysfs_add_one()
>>
>>     kobject_add failed for 1023 with -EEXIST, don't try to register
>>     things with the same name in the same directory.
>>
>> On my system, hotplug slots themselves can be added, removed and
>> replaced with the ohter type of I/O box. The ACPI firmware tells OS
>> the presence of those slots using _STA method (That is, it doesn't
>> use 'LoadTable()' AML operator). On the other hand, current pci_slot
>> driver doesn't check _STA. As a result, pci_slot driver tryied to
>> register the invalid (non-existing) slots. The ACPI firmware of my
>> system returns '1023' if the invalid slot's _SUN is evaluated. This
>> is the cause of Call Traces mentioned above. To fix this problem,
>> pci_slot driver need to check _STA when scanning ACPI Namespace.
>>
>> I'm sorry for reporting this so late. I'm attaching the patch to fix
>> the problem. This is against 2.6.24-rc3 with your patches applied.
>> Could you try it?
>>
>> BTW, acpiphp also seems to have the same problem...
>>
>> Thanks,
>> Kenji Kaneshige
> 
> Hello - back to this mail for a moment.  I wanted to talk about acpiphp
> again.  You say that acpiphp has this same problem as Alex's driver wrt
> evalating _SUN on a non existant object.  If we skip registering the
> slot in acpiphp's register_slot function if it returns 0 for _STA, we
> will not detect any hotplug slots, since leaf nodes will return 0 for
> _STA if a device is not present. Should we only skip it if it's parent
> is a p2p bridge that has _EJ0 and the parent's _STA says empty?
> 
> Your input is appreciated,
> Kristen
> 
>> ---
>>  drivers/acpi/pci_slot.c |   13 +++++++++++++
>>  1 file changed, 13 insertions(+)
>>
>> Index: linux-2.6.24-rc3/drivers/acpi/pci_slot.c
>> ===================================================================
>> --- linux-2.6.24-rc3.orig/drivers/acpi/pci_slot.c
>> +++ linux-2.6.24-rc3/drivers/acpi/pci_slot.c
>> @@ -113,10 +113,17 @@ register_slot(acpi_handle handle, u32 lv
>>  	int device;
>>  	unsigned long sun;
>>  	char name[KOBJ_NAME_LEN];
>> +	acpi_status status;
>> +	struct acpi_device *dummy_device;
>>  
>>  	struct pci_slot *pci_slot;
>>  	struct pci_bus *pci_bus = context;
>>  
>> +	/* Skip non-existing device object. */
>> +	status = acpi_bus_get_device(handle, &dummy_device);
>> +	if (ACPI_FAILURE(status))
>> +		return AE_OK;
>> +
>>  	if (check_slot(handle, &device, &sun))
>>  		return AE_OK;
>>  
>> @@ -150,12 +157,18 @@ walk_p2p_bridge(acpi_handle handle, u32 
>>  	acpi_status status;
>>  	acpi_handle dummy_handle;
>>  	acpi_walk_callback user_function;
>> +	struct acpi_device *dummy_device;
>>  
>>  	struct pci_dev *dev;
>>  	struct pci_bus *pci_bus;
>>  	struct p2p_bridge_context child_context;
>>  	struct p2p_bridge_context *parent_context = context;
>>  
>> +	/* Skip non-existing device object. */
>> +	status = acpi_bus_get_device(handle, &dummy_device);
>> +	if (ACPI_FAILURE(status))
>> +		return AE_OK;
>> +
>>  	pci_bus = parent_context->pci_bus;
>>  	user_function = parent_context->user_function;
>>  
>>
> 
>