alpha: fix RTC on marvel

Unlike other alphas, marvel doesn't have real PC-style CMOS clock hardware
- RTC accesses are emulated via PAL calls.  Unfortunately, for unknown
reason these calls work only on CPU #0.  So current implementation for
arbitrary CPU makes CMOS_READ/WRITE to be executed on CPU #0 via IPI.
However, for obvious reason this doesn't work with standard
get/set_rtc_time() functions, where a bunch of CMOS accesses is done with
disabled interrupts.

Solved by making the IPI calls for entire get/set_rtc_time() functions,
not for individual CMOS accesses.  Which is also a lot more effective
performance-wise.

The patch is largely based on the code from Jay Estabrook.
My changes:
- tweak asm-generic/rtc.h by adding a couple of #defines to
  avoid a massive code duplication in arch/alpha/include/asm/rtc.h;
- sys_marvel.c: fix get/set_rtc_time() return values (Jay's FIXMEs).

NOTE: this fixes *only* LIB_RTC drivers.  Legacy (CONFIG_RTC) driver
wont't work on marvel.  Actually I think that we should just disable
CONFIG_RTC on alpha (maybe in 2.6.30?), like most other arches - AFAIK,
all modern distributions use LIB_RTC anyway.

Signed-off-by: Ivan Kokshaysky <ink@jurassic.park.msu.ru>
Cc: Richard Henderson <rth@twiddle.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

1 files changed, 10 insertions, 0 deletions

diff --git a/arch/alpha/kernel/time.c b/arch/alpha/kernel/time.c
index e6a231435cba..b04e2cbf23a4 100644
--- a/arch/alpha/kernel/time.c
+++ b/arch/alpha/kernel/time.c
@@ -46,6 +46,7 @@
 #include <asm/io.h>
 #include <asm/hwrpb.h>
 #include <asm/8253pit.h>
+#include <asm/rtc.h>
 
 #include <linux/mc146818rtc.h>
 #include <linux/time.h>
@@ -180,6 +181,15 @@ common_init_rtc(void)
         init_rtc_irq();
 }
 
+unsigned int common_get_rtc_time(struct rtc_time *time)
+{
+        return __get_rtc_time(time);
+}
+
+int common_set_rtc_time(struct rtc_time *time)
+{
+        return __set_rtc_time(time);
+}
 
 /* Validate a computed cycle counter result against the known bounds for
    the given processor core.  There's too much brokenness in the way of