Bruce Momjian wrote:
> Prompted by confusion over Itanium/Opterion, I have written a patch to
> improve the way we define spinlocks for platforms and cpu's. It
> basically decouples the OS from the CPU spinlock code. In almost all
> cases, the spinlock code cares only about the compiler and CPU, not the
> OS.
>
> The patch:
>
> o defines HAS_TEST_AND_SET inside each spinlock routine, not in
> platform-specific files
> o moves slock_t defines into the spinlock routines
> o remove NEED_{CPU}_TAS_ASM define because it is no longer needed
> o reports a compile error if spinlocks are not defined
> o adds a configure option --without-spinlocks to allow
> non-spinlock compiles
OK, we have to decide which parts of this patch we want added. I think
there was agreement that we want this part for 7.4:
> o reports a compile error if spinlocks are not defined
> o adds a configure option --without-spinlocks to allow
> non-spinlock compiles
Now, do we also want to centralize the cpu tests in s_lock.h, or try to
patch up the include/port/*.h files for Opteron/Itanium. If you look in
freebsd.h, for example, you will see we basically try to duplicate the
cpu tests done in s_lock.h. To get this working, we would need to add
Itanium/Opteron tests there, and add i386 tests in other platforms that
previously supported only i386 and add Opteron/Itanium.
Or we could just apply the entire patch.
I am going to get some folks to test the patch today on a few platforms
to see how it works for them.
-- Bruce Momjian | http://candle.pha.pa.us pgman@candle.pha.pa.us | (610)
359-1001+ If your life is a hard drive, | 13 Roberts Road + Christ can be your backup. | Newtown Square,
Pennsylvania19073
