Source-Changes-HG archive
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index][Old Index]
[src/trunk]: src/sys/arch/aarch64 * avoid deadlock. mutex_owned() works only ...
details: https://anonhg.NetBSD.org/src/rev/ceea23e3255b
branches: trunk
changeset: 324228:ceea23e3255b
user: ryo <ryo%NetBSD.org@localhost>
date: Sat Jul 21 13:08:35 2018 +0000
description:
* avoid deadlock. mutex_owned() works only for adaptive lock, therefore we cannot use it for spinlock...
* add more NULL check
* clear pte when pmap_enter() fails
diffstat:
sys/arch/aarch64/aarch64/pmap.c | 97 ++++++++++++++++++++++++++++------------
sys/arch/aarch64/include/cpu.h | 5 +-
2 files changed, 72 insertions(+), 30 deletions(-)
diffs (255 lines):
diff -r 1eb4b7a9399e -r ceea23e3255b sys/arch/aarch64/aarch64/pmap.c
--- a/sys/arch/aarch64/aarch64/pmap.c Sat Jul 21 09:51:20 2018 +0000
+++ b/sys/arch/aarch64/aarch64/pmap.c Sat Jul 21 13:08:35 2018 +0000
@@ -1,4 +1,4 @@
-/* $NetBSD: pmap.c,v 1.10 2018/07/17 09:58:14 ryo Exp $ */
+/* $NetBSD: pmap.c,v 1.11 2018/07/21 13:08:35 ryo Exp $ */
/*
* Copyright (c) 2017 Ryo Shimizu <ryo%nerv.org@localhost>
@@ -27,7 +27,7 @@
*/
#include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: pmap.c,v 1.10 2018/07/17 09:58:14 ryo Exp $");
+__KERNEL_RCSID(0, "$NetBSD: pmap.c,v 1.11 2018/07/21 13:08:35 ryo Exp $");
#include "opt_arm_debug.h"
#include "opt_ddb.h"
@@ -149,6 +149,10 @@
#define LX_BLKPAG_ATTR_DEVICE_MEM __SHIFTIN(3, LX_BLKPAG_ATTR_INDX)
#define LX_BLKPAG_ATTR_MASK LX_BLKPAG_ATTR_INDX
+/* flags for cpu_info->ci_pmap_flags */
+#define CI_PMAP_FLAGS_KPM_OWNLOCKED 0x00000001
+
+
struct pv_entry {
TAILQ_ENTRY(pv_entry) pv_link;
struct pmap *pv_pmap;
@@ -189,16 +193,36 @@
mutex_exit(&md->mdpg_pvlock);
}
-#define PM_LOCKED(pm) \
- mutex_owned(&(pm)->pm_lock)
-#define PM_LOCK(pm) \
- do { \
- mutex_enter(&(pm)->pm_lock); \
- } while (0 /*CONSTCOND*/)
-#define PM_UNLOCK(pm) \
- do { \
- mutex_exit(&(pm)->pm_lock); \
- } while (0 /*CONSTCOND*/)
+
+static inline int
+pm_kpm_ownlocked(struct pmap *pm)
+{
+ if (pm != pmap_kernel())
+ return false;
+
+ return curcpu()->ci_pmap_flags & CI_PMAP_FLAGS_KPM_OWNLOCKED;
+}
+
+static inline void
+pm_lock(struct pmap *pm)
+{
+ mutex_enter(&pm->pm_lock);
+
+ /*
+ * remember locking myself if pm is pmap_kernel.
+ * mutex_owned() works only for adaptive lock, therefore we cannot use.
+ */
+ if (pm == pmap_kernel())
+ curcpu()->ci_pmap_flags |= CI_PMAP_FLAGS_KPM_OWNLOCKED;
+}
+
+static inline void
+pm_unlock(struct pmap *pm)
+{
+ if (pm == pmap_kernel())
+ curcpu()->ci_pmap_flags &= ~CI_PMAP_FLAGS_KPM_OWNLOCKED;
+ mutex_exit(&pm->pm_lock);
+}
static void __unused
pm_addr_check(struct pmap *pm, vaddr_t va, const char *prefix)
@@ -881,7 +905,8 @@
pmap_pv_unlock(md);
- pool_cache_put(&_pmap_pv_pool, pv);
+ if (pv != NULL)
+ pool_cache_put(&_pmap_pv_pool, pv);
}
#if defined(PMAP_PV_DEBUG) || defined(DDB)
@@ -1055,7 +1080,7 @@
mdattr = VM_PAGE_TO_MD(pg)->mdpg_flags &
(VM_PROT_READ | VM_PROT_WRITE);
- PM_LOCK(pv->pv_pmap);
+ pm_lock(pv->pv_pmap);
ptep = pv->pv_ptep;
pte = *ptep;
@@ -1079,7 +1104,7 @@
aarch64_tlbi_by_va(pv->pv_va);
#endif
- PM_UNLOCK(pv->pv_pmap);
+ pm_unlock(pv->pv_pmap);
}
void
@@ -1109,7 +1134,7 @@
KDASSERT((sva & PAGE_MASK) == 0);
KDASSERT((eva & PAGE_MASK) == 0);
- PM_LOCK(pm);
+ pm_lock(pm);
for (va = sva; va < eva; va += PAGE_SIZE) {
pt_entry_t *ptep, pte;
@@ -1162,7 +1187,7 @@
}
}
- PM_UNLOCK(pm);
+ pm_unlock(pm);
}
void
@@ -1335,9 +1360,9 @@
* _pmap_enter() may be called recursively. In case of
* pmap_enter() -> _pmap_enter_pv() -> pool_cache_get() -> pmap_kenter()
*/
- pm_locked = PM_LOCKED(pm);
+ pm_locked = pm_kpm_ownlocked(pm);
if (!pm_locked)
- PM_LOCK(pm);
+ pm_lock(pm);
/*
* traverse L0 -> L1 -> L2 -> L3 table with growing pdp if needed.
@@ -1422,13 +1447,27 @@
*/
mdattr = VM_PROT_READ | VM_PROT_WRITE;
if (pg != NULL) {
+ if (pm != pmap_kernel()) {
+ /* avoid deadlock (see above comment) */
+ pm_lock(pmap_kernel());
+ }
error = _pmap_enter_pv(pg, pm, va, ptep, pa, flags);
+ if (pm != pmap_kernel()) {
+ pm_unlock(pmap_kernel());
+ }
+
if (error != 0) {
+ /*
+ * If pmap_enter() fails,
+ * it must not leave behind an existing pmap entry.
+ */
+ if (!kenter && ((pte & LX_BLKPAG_OS_WIRED) == 0))
+ atomic_swap_64(ptep, 0);
+
PMAP_COUNT(pv_entry_cannotalloc);
if (flags & PMAP_CANFAIL)
goto done;
- panic(
- "pmap_enter: failed to allocate pv_entry");
+ panic("pmap_enter: failed to allocate pv_entry");
}
/* update referenced/modified flags */
@@ -1477,7 +1516,7 @@
done:
if (!pm_locked)
- PM_UNLOCK(pm);
+ pm_unlock(pm);
return error;
}
@@ -1509,7 +1548,7 @@
UVMHIST_LOG(pmaphist, "pm=%p, va=%016lx, kremovemode=%d",
pm, va, kremove, 0);
- PM_LOCK(pm);
+ pm_lock(pm);
ptep = _pmap_pte_lookup(pm, va);
if (ptep != NULL) {
@@ -1539,7 +1578,7 @@
pm->pm_stats.resident_count--;
}
done:
- PM_UNLOCK(pm);
+ pm_unlock(pm);
}
void
@@ -1618,7 +1657,7 @@
PM_ADDR_CHECK(pm, va);
- PM_LOCK(pm);
+ pm_lock(pm);
ptep = _pmap_pte_lookup(pm, va);
if (ptep != NULL) {
pte = *ptep;
@@ -1626,7 +1665,7 @@
((pte & LX_BLKPAG_OS_WIRED) == 0)) {
/* invalid pte, or pte is not wired */
PMAP_COUNT(unwire_failure);
- PM_UNLOCK(pm);
+ pm_unlock(pm);
return;
}
@@ -1635,7 +1674,7 @@
pm->pm_stats.wired_count--;
}
- PM_UNLOCK(pm);
+ pm_unlock(pm);
}
bool
@@ -1670,7 +1709,7 @@
}
#endif
- PM_LOCK(pm);
+ pm_lock(pm);
ptep = _pmap_pte_lookup(pm, va);
if (ptep == NULL) {
@@ -1778,7 +1817,7 @@
fixed = true;
done:
- PM_UNLOCK(pm);
+ pm_unlock(pm);
return fixed;
}
diff -r 1eb4b7a9399e -r ceea23e3255b sys/arch/aarch64/include/cpu.h
--- a/sys/arch/aarch64/include/cpu.h Sat Jul 21 09:51:20 2018 +0000
+++ b/sys/arch/aarch64/include/cpu.h Sat Jul 21 13:08:35 2018 +0000
@@ -1,4 +1,4 @@
-/* $NetBSD: cpu.h,v 1.3 2018/07/09 06:19:53 ryo Exp $ */
+/* $NetBSD: cpu.h,v 1.4 2018/07/21 13:08:35 ryo Exp $ */
/*-
* Copyright (c) 2014 The NetBSD Foundation, Inc.
@@ -78,6 +78,9 @@
struct evcnt ci_vfp_reuse;
struct evcnt ci_vfp_save;
struct evcnt ci_vfp_release;
+
+ /* per cpu pmap private */
+ uint64_t ci_pmap_flags;
};
static inline struct cpu_info *
Home |
Main Index |
Thread Index |
Old Index