[src/trunk]: src/sys/arch/sparc64/sparc64 since pmap_activate/pmap_deactivate...

[mrg <mrg%NetBSD.org@localhost>]

details:   https://anonhg.NetBSD.org/src/rev/7a745f61e2ce
branches:  trunk
changeset: 753262:7a745f61e2ce
user:      mrg <mrg%NetBSD.org@localhost>
date:      Sun Mar 21 22:38:08 2010 +0000

description:
since pmap_activate/pmap_deactivate need to not block, but the rest of
pmap is OK, add a new IPL_VM pmap_ctx_lock that is held while allocating
or freeing a context, and revert pmap_lock back to IPL_NONE.

i considered a per-cpu context lock, but that would require locking
each active CPUs lock in turn in pmap_remove_all() and pmap_destroy().


thanks to ad for explaining how these parts interact.

diffstat:

 sys/arch/sparc64/sparc64/pmap.c |  22 ++++++++++++----------
 1 files changed, 12 insertions(+), 10 deletions(-)

diffs (99 lines):

diff -r be1b7e29a618 -r 7a745f61e2ce sys/arch/sparc64/sparc64/pmap.c
--- a/sys/arch/sparc64/sparc64/pmap.c   Sun Mar 21 21:23:51 2010 +0000
+++ b/sys/arch/sparc64/sparc64/pmap.c   Sun Mar 21 22:38:08 2010 +0000
@@ -1,4 +1,4 @@
-/*     $NetBSD: pmap.c,v 1.260 2010/03/20 20:36:23 mrg Exp $   */
+/*     $NetBSD: pmap.c,v 1.261 2010/03/21 22:38:08 mrg Exp $   */
 /*
  *
  * Copyright (C) 1996-1999 Eduardo Horvath.
@@ -26,7 +26,7 @@
  */
 
 #include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: pmap.c,v 1.260 2010/03/20 20:36:23 mrg Exp $");
+__KERNEL_RCSID(0, "$NetBSD: pmap.c,v 1.261 2010/03/21 22:38:08 mrg Exp $");
 
 #undef NO_VCACHE /* Don't forget the locked TLB in dostart */
 #define        HWREF
@@ -330,6 +330,7 @@
  * Global pmap lock.
  */
 static kmutex_t pmap_lock;
+static kmutex_t pmap_ctx_lock;
 static bool lock_available = false;
 
 /*
@@ -1293,7 +1294,8 @@
        vm_first_phys = avail_start;
        vm_num_phys = avail_end - avail_start;
 
-       mutex_init(&pmap_lock, MUTEX_SPIN, IPL_VM);
+       mutex_init(&pmap_lock, MUTEX_DEFAULT, IPL_NONE);
+       mutex_init(&pmap_ctx_lock, MUTEX_SPIN, IPL_VM);
 #if defined(USE_LOCKSAFE_PSEG_GETSET)
        mutex_init(&pseg_lock, MUTEX_SPIN, IPL_VM);
 #endif
@@ -1415,7 +1417,7 @@
                return;
        }
        DPRINTF(PDB_DESTROY, ("pmap_destroy: freeing pmap %p\n", pm));
-       mutex_enter(&pmap_lock);
+       mutex_enter(&pmap_ctx_lock);
 #ifdef MULTIPROCESSOR
        for (ci = cpus; ci != NULL; ci = ci->ci_next) {
                if (CPUSET_HAS(cpus_active, ci->ci_index))
@@ -1424,7 +1426,7 @@
 #else
        ctx_free(pm, curcpu());
 #endif
-       mutex_exit(&pmap_lock);
+       mutex_exit(&pmap_ctx_lock);
 
        /* we could be a little smarter and leave pages zeroed */
        for (pg = TAILQ_FIRST(&pm->pm_obj.memq); pg != NULL; pg = nextpg) {
@@ -1922,7 +1924,7 @@
        write_user_windows();
        pm->pm_refs = 0;
 
-       mutex_enter(&pmap_lock);
+       mutex_enter(&pmap_ctx_lock);
 #ifdef MULTIPROCESSOR
        CPUSET_CLEAR(pmap_cpus_active);
        for (ci = cpus; ci != NULL; ci = ci->ci_next) {
@@ -1935,7 +1937,7 @@
 #else
        ctx_free(pm, curcpu());
 #endif
-       mutex_exit(&pmap_lock);
+       mutex_exit(&pmap_ctx_lock);
 
        REMOVE_STAT(flushes);
 #ifdef MULTIPROCESSOR
@@ -3071,7 +3073,7 @@
 
        KASSERT(pm != pmap_kernel());
        KASSERT(pm == curproc->p_vmspace->vm_map.pmap);
-       mutex_enter(&pmap_lock);
+       mutex_enter(&pmap_ctx_lock);
        ctx = curcpu()->ci_pmap_next_ctx++;
 
        /*
@@ -3106,7 +3108,7 @@
        curcpu()->ci_ctxbusy[ctx] = pm->pm_physaddr;
        LIST_INSERT_HEAD(&curcpu()->ci_pmap_ctxlist, pm, pm_list[cpu_number()]);
        pmap_ctx(pm) = ctx;
-       mutex_exit(&pmap_lock);
+       mutex_exit(&pmap_ctx_lock);
        DPRINTF(PDB_CTX_ALLOC, ("ctx_alloc: cpu%d allocated ctx %d\n",
                cpu_number(), ctx));
        return ctx;
@@ -3121,7 +3123,7 @@
        int oldctx;
        int cpunum;
 
-       KASSERT(mutex_owned(&pmap_lock));
+       KASSERT(mutex_owned(&pmap_ctx_lock));
 
 #ifdef MULTIPROCESSOR
        cpunum = ci->ci_index;