Fix RTC wakealarm sysfs interface breakage.

Commit ed458df4d2470adc02762a87a9ad665d0b1a2bd4 ("PnP: move
pnpacpi/pnpbios_init to after PCI init") moved the PnP RTC discovery
later, and now the ACPI RTC glue code doesn't find it any more, breaking
the RTC wakealarm sysfs interfaces, as reported by Rafael.

This really is fairly messy, and we have several annoying ordering
constraints here - the PnP code that sets up the RTC resources wants to
run after the PCI resources have to be registered, which in turn needs
to run after ACPI has at least enumerated the root PCI buses etc.  Our
initcall ordering is not fine-grained enough to make this all painless.

So this moves the ACPI RTC glue ("acpi_rtc_init()") down to a regular
module call, which fixes the problem Rafael has.  The reason this isn't
wonderful is that we really should do acpi_rtc_init before we do the
rtc_cmos init, and now those two are in the same module_init() section.

Which happens to work, but only because drivers/rtc is linked after
drivers/acpi.  In other words, we still have a very subtle ordering
issue here. Grr.

Reported-and-tested-by: Rafael J. Wysocki <rjw@sisk.pl>
Acked-by: David Brownell <david-b@pacbell.net>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

diff --git a/drivers/acpi/glue.c b/drivers/acpi/glue.c
index 8dd3336efd7e242209240010b7fa9710bb2ae6e8..3c578ef78c4830f791039b83e2a44ab80d7538d8 100644 (file)
--- a/drivers/acpi/glue.c
+++ b/drivers/acpi/glue.c
@@ -369,7 +369,6 @@ static int __init acpi_rtc_init(void)
                DBG("RTC unavailable?\n");
        return 0;
 }
-/* do this between RTC subsys_initcall() and rtc_cmos driver_initcall() */
-fs_initcall(acpi_rtc_init);
+module_init(acpi_rtc_init);
 
 #endif