On Fri, 6 Sep 2019, Shevchenko, Andriy wrote:
> On Wed, Sep 04, 2019 at 09:20:03AM +0200, Nikolaus Voss wrote:
>> On Fri, 30 Aug 2019, Shevchenko, Andriy wrote:
>>> On Fri, Aug 16, 2019 at 01:57:26PM +0200, Nikolaus Voss wrote:
>>>> On Wed, 14 Aug 2019, Schmauss, Erik wrote:
>>>>>> -----Original Message-----
>>>>>> From: Shevchenko, Andriy
>>>>>> Sent: Wednesday, August 14, 2019 11:51 AM
>>>>>> To: Nikolaus Voss <[email protected]>
>>>>>> Cc: Rafael J. Wysocki <[email protected]>; Len Brown <[email protected]>;
>>>>>> Moore, Robert <[email protected]>; Schmauss, Erik
>>>>>> <[email protected]>; Jacek Anaszewski <[email protected]>;
>>>>>> Pavel Machek <[email protected]>; Dan Murphy <[email protected]>; Thierry
>>>>>> Reding <[email protected]>; [email protected];
>>>>>> [email protected]; [email protected]; [email protected]
>>>>>> Subject: Re: [PATCH 1/3] ACPI: Resolve objects on host-directed table loads
>>>>>>
>>>>>> On Wed, May 29, 2019 at 02:18:20PM +0200, Nikolaus Voss wrote:
>>>>>>> If an ACPI SSDT overlay is loaded after built-in tables have been
>>>>>>> loaded e.g. via configfs or efivar_ssdt_load() it is necessary to
>>>>>>> rewalk the namespace to resolve references. Without this, relative and
>>>>>>> absolute paths like ^PCI0.SBUS or \_SB.PCI0.SBUS are not resolved
>>>>>>> correctly.
>>>>>>>
>>>>>>> Make configfs load use the same method as efivar_ssdt_load().
>>>>>>
>>>>>> This patch brought a regression (bisect log below).
>>>>>> Now I'm unable to unload the table which was working before.
>>>>>>
>>>>>> Reverting (manual, due to ACPICA changes) helps.
>>>>>>
>>>>>> Please, consider to revert for this cycle, or fix. I will be glad to test any
>>>>>> proposed fix.
>>>>>
>>>>> We submitted a patch (d1fb5b2f623b1af5a0d2a83d205df1b61f430dc6)
>>>>> in response to this suggestion and I was not aware that this had been applied.
>>>>>
>>>>> Rafael, please revert at least the ACPICA portion of this patch.
>>>>
>>>> As I see it, my ACPICA change is not part of 5.3-rc1 any more. Reverting my
>>>> fix is part of the patch above (d1fb5b2f623b1af5a0d2a83d205df1b61f430dc6)
>>>> which is already applied.
>>>>
>>>> Nevertheless, what is new, is that acpi_ns_initialize_objects() is called in
>>>> acpi_load_table(). This is necessary to resolve the references in the newly
>>>> loaded table. Maybe this prevents the table from being unloaded?
>>>
>>> So, can we do something about it? It's a regression.
>>>
>>> Rafael, Nikolaus?
>>
>> can you describe how you unload the table (from userspace?). I cannot
>> reproduce this regression. I was not aware of any working interface for
>> unloading ACPI tables, I ended up in kexec'ing the kernel for my tests each
>> time I had to unload a table.
>
> Sure.
>
> I have connected an I?C device(s) to my board where I have compiled ACPI tables
> which describes them (more details if you want to know is on [1]).
>
> So, I create a folder in ConfigFS [1,2] and fill it up with SSDT (an *.aml file).
> After this, if I try to remove the folder in ConfigFS followed by table removal
> event, the actual nodes won't be removed, and this messes up with the internal
> representation of the ACPI device tree.
>
> [1]: https://www.kernel.org/doc/html/latest/admin-guide/acpi/ssdt-overlays.html
> [2]: https://htot.github.io/meta-intel-edison/1.3-ACPI-or-not.html#run-time-loading-through-configfs
Thanks, I have reproduced it and the culprit is my acpi_configfs patch
that now uses acpi_load_table() to have the references resolved.
My proposal would be to have that function return the table index, which
is needed to unload the table. I'll submit a patch, Andy, could you test
it in your setup?
Erik/Robert, is it ok to change the acpi_load_table() ACPICA API
function? It has only a few uses.
Niko