On Tue, Mar 1, 2016 at 9:36 AM, Toshi Kani [off-list ref] wrote:

On Tue, 2016-03-01 at 16:03 +0000, Moore, Robert wrote:

quoted

We have a bunch of macros in include/acmacros.h -- like this:

ACPI_MOVE_16_TO_16(d, s)

There is a problem in using the ACPICA byte-swap macros.  ACPI is little-
endian arch, so the macros are set to perform byte-swappings when the CPU
arch is big-endian.  This case, however, is the other way around.  The
fields in question are defined & stored as arrays of bytes.  If you treat
them as multi-bytes numeric values, then you need to byte-swap them when
the CPU arch is little-endian because arrays of bytes have the same
addressing as big-endian.

Another issue is that it is not clear who needs to perform the byte-
swapping among ACPICA and drivers.  If ACPICA, drivers must agree that
these fields are always treated as multi-bytes numeric values despite of
the spec.  If drivers, we need to make sure that only a single driver
performs this byte-swapping one time as ACPI tables are global structures.

I think it is much clearer to define the structure according to the ACPI
spec.

I think the "ACPI tables are little-endian" assumption is pervasive
throughout the implementation.

Toshi, it seems all we need is conversions like:

- sprintf(buf, "%#x\n", dcr->vendor_id);
+ sprintf(buf, "%#x\n", le16_to_cpu(dcr->vendor_id));

...for the values exported to userspace through sysfs, but otherwise
leave the base table definitions as is.  Will this suffice?