[oe] [meta-java][PATCH 1/1] openjdk: fix a random crash on JNI_CreateJavaVM

Wenlin Kang wenlin.kang at windriver.com
Tue Feb 19 02:30:57 UTC 2019


When call JNI_CreateJavaVM API, sometimes it will crash(we have
reproduced it on quemuppc), this patch fixes it.

Signed-off-by: Wenlin Kang <wenlin.kang at windriver.com>
---
 .../openjdk-8-release-172b11-common.inc       |  1 +
 ...otspot-fix-crash-on-JNI_CreateJavaVM.patch | 36 +++++++++++++++++++
 2 files changed, 37 insertions(+)
 create mode 100644 recipes-core/openjdk/patches-openjdk-8/0001-hotspot-fix-crash-on-JNI_CreateJavaVM.patch

diff --git a/recipes-core/openjdk/openjdk-8-release-172b11-common.inc b/recipes-core/openjdk/openjdk-8-release-172b11-common.inc
index ad20585..b37608c 100644
--- a/recipes-core/openjdk/openjdk-8-release-172b11-common.inc
+++ b/recipes-core/openjdk/openjdk-8-release-172b11-common.inc
@@ -21,6 +21,7 @@ PATCHES_URI = "\
     ${LIBPNG_NEON_PATCH} \
     file://handle_extra_output.patch \
     file://hotspot-handle-gcc7-format-overflow.patch \
+    file://0001-hotspot-fix-crash-on-JNI_CreateJavaVM.patch \
 "
 HOTSPOT_UB_PATCH = "\
     file://0014-hotspot-zero-fix-undefined-behaviour-gcc-v8-fix.patch \
diff --git a/recipes-core/openjdk/patches-openjdk-8/0001-hotspot-fix-crash-on-JNI_CreateJavaVM.patch b/recipes-core/openjdk/patches-openjdk-8/0001-hotspot-fix-crash-on-JNI_CreateJavaVM.patch
new file mode 100644
index 0000000..4618b6b
--- /dev/null
+++ b/recipes-core/openjdk/patches-openjdk-8/0001-hotspot-fix-crash-on-JNI_CreateJavaVM.patch
@@ -0,0 +1,36 @@
+From d0ebd6f108df9b0f9533de6f16280ceec136d8e6 Mon Sep 17 00:00:00 2001
+From: Wenlin Kang <wenlin.kang at windriver.com>
+Date: Sun, 17 Feb 2019 22:38:33 -0800
+Subject: [PATCH] hotspot: fix crash on JNI_CreateJavaVM
+
+In function os::pd_create_stack_guard_pages(char* addr, size_t size),
+when addr < os::Linux::initial_thread_stack_bottom(), usually munmap()
+will not be called, but when mincore()==-1, get_stack_commited_bottom()
+will make stack_extent < os::Linux::initial_thread_stack_bottom() and
+stack_extent < addr too, then munmap() is called, in such case, it may
+cause segment(we have reproduced it on linux_PPC).
+
+Upstream-Status: Pending
+
+Signed-off-by: Wenlin Kang <wenlin.kang at windriver.com>
+---
+ hotspot/src/os/linux/vm/os_linux.cpp | 3 ++-
+ 1 file changed, 2 insertions(+), 1 deletion(-)
+
+diff --git a/hotspot/src/os/linux/vm/os_linux.cpp b/hotspot/src/os/linux/vm/os_linux.cpp
+index 044a70a6..b8246ce4 100644
+--- a/hotspot/src/os/linux/vm/os_linux.cpp
++++ b/hotspot/src/os/linux/vm/os_linux.cpp
+@@ -3078,7 +3078,8 @@ bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
+     uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
+     unsigned char vec[1];
+ 
+-    if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
++    if ((mincore((address)stack_extent, os::vm_page_size(), vec) == -1)
++        && ((size_t)addr > (size_t)stack_extent)) {
+       // Fallback to slow path on all errors, including EAGAIN
+       stack_extent = (uintptr_t) get_stack_commited_bottom(
+                                     os::Linux::initial_thread_stack_bottom(),
+-- 
+2.17.1
+
-- 
2.17.1



More information about the Openembedded-devel mailing list