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 >>>>>> Cc: Rafael J. Wysocki ; Len Brown ; >>>>>> Moore, Robert ; Schmauss, Erik >>>>>> ; Jacek Anaszewski ; >>>>>> Pavel Machek ; Dan Murphy ; Thierry >>>>>> Reding ; linux-acpi@vger.kernel.org; >>>>>> devel@acpica.org; linux-leds@vger.kernel.org; linux-pwm@vger.kernel.org >>>>>> 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 I²C 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