On Thu, Mar 5, 2015 at 11:17 AM, Will Deacon [off-list ref] wrote:

On Wed, Mar 04, 2015 at 11:14:50PM +0000, Rafael J. Wysocki wrote:

quoted

On Monday, March 02, 2015 12:00:21 PM Al Stone wrote:

quoted

On 03/02/2015 10:29 AM, Will Deacon wrote:

quoted

On Wed, Feb 25, 2015 at 12:36:24AM +0000, al.stone at linaro.org wrote:

quoted

diff --git a/arch/arm64/kernel/acpi-blacklist.c b/arch/arm64/kernel/acpi-blacklist.c
new file mode 100644
index 0000000..1be6a56
--- /dev/null
+++ b/arch/arm64/kernel/acpi-blacklist.c

@@ -0,0 +1,20 @@
+/*
+ *  ARM64 Specific ACPI Blacklist Support
+ *
+ *  Copyright (C) 2015, Linaro Ltd.
+ *      Author: Al Stone <al.stone@linaro.org>
+ *
+ *  This program is free software; you can redistribute it and/or modify
+ *  it under the terms of the GNU General Public License version 2 as
+ *  published by the Free Software Foundation.
+ */
+
+#define pr_fmt(fmt) "ACPI: " fmt
+
+#include <linux/acpi.h>
+
+/* The arm64 ACPI blacklist is currently empty.  */
+int __init acpi_blacklisted(void)
+{
+        return 0;
+}

diff --git a/arch/arm64/kernel/acpi-osi.c b/arch/arm64/kernel/acpi-osi.c
new file mode 100644
index 0000000..bb351f4
--- /dev/null
+++ b/arch/arm64/kernel/acpi-osi.c

@@ -0,0 +1,25 @@
+/*
+ *  ARM64 Specific ACPI _OSI Support
+ *
+ *  Copyright (C) 2015, Linaro Ltd.
+ *      Author: Al Stone <al.stone@linaro.org>
+ *
+ *  This program is free software; you can redistribute it and/or modify
+ *  it under the terms of the GNU General Public License version 2 as
+ *  published by the Free Software Foundation.
+ */
+
+#define pr_fmt(fmt) "ACPI: " fmt
+
+#include <linux/acpi.h>
+
+/*
+ * Consensus is to deprecate _OSI for all new ACPI-supported architectures.
+ * So, for arm64, reduce _OSI to a warning message, and tell the firmware
+ * nothing of value.
+ */
+u32 acpi_osi_handler(acpi_string interface, u32 supported)
+{
+        pr_warn("_OSI was called, but is deprecated for this architecture.\n");
+        return false;
+}

This kinda feels backwards to me. If _OSI is going away, then the default
should be "the architecture doesn't need to do anything", rather than have
new architectures defining a bunch of empty, useless stub code.

Anyway we could make this the default in core code and have architectures
that *do* want _OSI override that behaviour, instead of the other way around?

We could do that; I personally don't have a strong preference either way,
so I'm inclined to make it whatever structure Rafael thinks is proper since
it affects ACPI code most.  That being said, the current patch structure
made sense to me since it wasn't distorting existing code much -- and given
the pure number of x86/ia64 machines vs ARM machines using ACPI, that seemed
the more cautious approach.

@Rafael: do you have an opinion/preference?

My preference is to avoid changes in the existing code at least for the time
being.  Especially if the changes in question are going to affect ia64, unless
you have an Itanium machine where you can readily test those, that is. :-)

Well, this code doesn't even need to compiled for ia64 if we have those
architectures that want to use _OSI select a Kconfig symbol for it, so I
don't think the testing argument is really that valid. I appreciate that you
want to avoid changing the existing code, but I also don't want to add this
sort of stuff to the architecture code, when it really has nothing to do
with the architecture.

OK, so consider this.

_OSI may be deprecated in the spec for *new* implementations.

However, there still are many systems out there that use _OSI and
we'll need to support
them going forward.  So while the spec people may think that they have
deprecated
_OSI, the reality is that in the kernel it is not going to be
deprecated as long as there
are systems using it that we need to support.

So the whole "_OSI is going away" argument is simply bogus and useless.

That aside, yes, we can use a Kconfig symbol to select from x86 and ia64
and compile the generic code conditional on that.  That would be fine by me.

Rafael