On Sun, Jul 19, 2015 at 06:33:29PM +0200, Mark Kettenis wrote:
> So my "unlocking the reaper" diff that was committed the other day
> breaks the MP architectures that don't have an "mpsafe" pmap yet (see
> my commit message). Is seems hppa's pmap is actually safe enough, but
> alpha, m88k, mips64 and powerpc aren't. The diff below fixes alpha
> and powerpc with a big hammer. Theo is currently building snapshots
> with this diff.
>
> visa@, you'll probably want to add something similar.
>
> miod@, mpi@, do you want me to move ahead and commit this temporary fix?
Here is a fix for mips64 pmap_remove(). There is a stat_count() (debug
only) outside the kernel lock so the increment should be done as an
atomic operation. Tested on MP octeon.
ok?
Index: arch/mips64/mips64/pmap.c
===================================================================
RCS file: src/sys/arch/mips64/mips64/pmap.c,v
retrieving revision 1.79
diff -u -p -r1.79 pmap.c
--- arch/mips64/mips64/pmap.c 2 May 2015 14:33:19 -0000 1.79
+++ arch/mips64/mips64/pmap.c 22 Jul 2015 13:20:21 -0000
@@ -123,7 +123,7 @@ struct {
if (pmapdebug & (flag)) \
printf printdata;
-#define stat_count(what) (what)++
+#define stat_count(what) atomic_inc_int(&(what))
int pmapdebug = PDB_ENTER|PDB_FOLLOW;
#else
@@ -649,6 +649,7 @@ pmap_remove(pmap_t pmap, vaddr_t sva, va
if (pmap == NULL)
return;
+ KERNEL_LOCK();
if (pmap == pmap_kernel()) {
/* remove entries from kernel pmap */
#ifdef DIAGNOSTIC
@@ -676,6 +677,7 @@ pmap_remove(pmap_t pmap, vaddr_t sva, va
pmap_invalidate_kernel_page(sva);
stat_count(remove_stats.flushes);
}
+ KERNEL_UNLOCK();
return;
}
@@ -718,6 +720,7 @@ pmap_remove(pmap_t pmap, vaddr_t sva, va
stat_count(remove_stats.flushes);
}
}
+ KERNEL_UNLOCK();
}
/*