Re: [RFC PATCH 02/10] acpi: install SSDT tables from initrd

[Octavian Purdila]

On Tue, Apr 5, 2016 at 3:49 AM, Zheng, Lv <lv.zheng@xxxxxxxxx> wrote:
> Hi,
>
>> From: Octavian Purdila [mailto:octavian.purdila@xxxxxxxxx]
>> Subject: Re: [RFC PATCH 02/10] acpi: install SSDT tables from initrd
>>
>> On Fri, Apr 1, 2016 at 1:11 PM, Octavian Purdila
>> <octavian.purdila@xxxxxxxxx> wrote:
>> > On Fri, Apr 1, 2016 at 8:05 AM, Zheng, Lv <lv.zheng@xxxxxxxxx> wrote:
>> >> Hi,
>> >>
>> >> IMO, there is already a similar function upstreamed:
>> >>
>> https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/commit/?id=c85c
>> c81
>> >> Could it work for your use case?
>> >
>> > Yes, it is basically the same.
>> >
>> > The only difference is on how we handle taint. I think we should use a
>> > new taint for overlays and that we don't need to disable lockdep.
>> >
>> > BTW, why is lockdep disabled when we override?
>>
>> The other thing I forgot to mention is that I think we should allow
>> installing new tables even if CONFIG_ACPI_INITRD_TABLE_OVERRIDE is not
>> selected. IMO the override and overlay functionality is different,
>> with the latter being more then a debug option.
> [Lv Zheng]
> I don't think so. The initrd override mechanism is not dependent on CONFIG_ACPI_DEBUG.
> According to the spec, we can allow a higher versioning same table (same table sig, table id, table oem id) to override the old tables as a patching functionality.
> So both the functionalities are not debug options and serve for the same purpose from this point of view.
> And IMO that's why the initrd override mechanism needn't be dependent on CONFIG_ACPI_DEBUG.
>

The problem is that CONFIG_ACPI_INITRD_TABLE_OVERRIDE is presented as
a debug option in Documentation/initrd_table_override.txt and most
distributions are not selecting it which makes it hard to use it in
practice.

> I'm really OK with removing the acpi_table_taint() for CONFIG_ACPI_INITRD_TABLE_OVERRIDE but leaving some info messages indicating the table upgrades.
> I don't think this mechanism is unsafe.
> It happens during a initialization step occurring before the table is loaded and hence should be safe even the synchronization is not so robust in ACPICA.
> And with the revision support added, we should be able to allow vendors to update the buggy tables.
> That means the tables may be originated from the safe sources - the vendors.
>
>>
>> I will prepare a patch for the next version of the series to decouple
>> installing new tables from CONFIG_ACPI_INITRD_TABLE_OVERRIDE.
> [Lv Zheng]
> I don't think they need to be decoupled.
> The use case is:
> If there is an ACPI table in initrd image and:
> 1. if the table's revision is higher than the existing one, override the existing one;
> 2. if the table is a brand new one, install it.
>

The implementation will stay the same of course, I was just suggesting
to move CONFIG_ACPI_INITRD_TABLE_OVERRIDE in
acpi_os_physical_table_override to allow new tables to be loaded even
if the option is not selected.