graal-jvmci-8: src/os/linux/vm/os

comparison src/os/linux/vm/os_linux.cpp @ 10969:a837fa3d3f86

8013057: assert(_needs_gc || SafepointSynchronize::is_at_safepoint()) failed: only read at safepoint Summary: Detect mmap() commit failures in Linux and Solaris os::commit_memory() impls and call vm_exit_out_of_memory(). Add os::commit_memory_or_exit(). Also tidy up some NMT accounting and some mmap() return value checking. Reviewed-by: zgu, stefank, dholmes, dsamersoff

author	dcubed
date	Thu, 13 Jun 2013 11:16:38 -0700
parents	f2110083203d
children	1f4355cee9a2

comparison

equal deleted inserted replaced

-:f2110083203d
+:a837fa3d3f86
 ::close(fd);
 unlink(buf);
 }
 }
+static bool recoverable_mmap_error(int err) {
+// See if the error is one we can let the caller handle. This
+// list of errno values comes from JBS-6843484. I can't find a
+// Linux man page that documents this specific set of errno
+// values so while this list currently matches Solaris, it may
+// change as we gain experience with this failure mode.
+switch (err) {
+case EBADF:
+case EINVAL:
+case ENOTSUP:
+// let the caller deal with these errors
+return true;
+default:
+// Any remaining errors on this OS can cause our reserved mapping
+// to be lost. That can cause confusion where different data
+// structures think they have the same memory mapped. The worst
+// scenario is if both the VM and a library think they have the
+// same memory mapped.
+return false;
+}
+}
+static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
+int err) {
+warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
+", %d) failed; error='%s' (errno=%d)", addr, size, exec,
+strerror(err), err);
+}
+static void warn_fail_commit_memory(char* addr, size_t size,
+size_t alignment_hint, bool exec,
+int err) {
+warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
+", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, size,
+alignment_hint, exec, strerror(err), err);
+}
 // NOTE: Linux kernel does not really reserve the pages for us.
 //       All it does is to check if there are enough free pages
 //       left at the time of mmap(). This could be a potential
 //       problem.
-bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
+int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
 if (res != (uintptr_t) MAP_FAILED) {
 if (UseNUMAInterleaving) {
 numa_make_global(addr, size);
 }
-return true;
+return 0;
 }
-return false;
+int err = errno;  // save errno from mmap() call above
+if (!recoverable_mmap_error(err)) {
+warn_fail_commit_memory(addr, size, exec, err);
+vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
+}
+return err;
+}
+bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
+return os::Linux::commit_memory_impl(addr, size, exec) == 0;
+}
+void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
+const char* mesg) {
+assert(mesg != NULL, "mesg must be specified");
+int err = os::Linux::commit_memory_impl(addr, size, exec);
+if (err != 0) {
+// the caller wants all commit errors to exit with the specified mesg:
+warn_fail_commit_memory(addr, size, exec, err);
+vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
+}
 }
 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
 #ifndef MAP_HUGETLB
 #define MAP_HUGETLB 0x40000
 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
 #ifndef MADV_HUGEPAGE
 #define MADV_HUGEPAGE 14
 #endif
-bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
+int os::Linux::commit_memory_impl(char* addr, size_t size,
-bool exec) {
+size_t alignment_hint, bool exec) {
+int err;
 if (UseHugeTLBFS && alignment_hint > (size_t)vm_page_size()) {
 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
 uintptr_t res =
 (uintptr_t) ::mmap(addr, size, prot,
 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS|MAP_HUGETLB,
 -1, 0);
 if (res != (uintptr_t) MAP_FAILED) {
 if (UseNUMAInterleaving) {
 numa_make_global(addr, size);
 }
-return true;
+return 0;
+}
+err = errno;  // save errno from mmap() call above
+if (!recoverable_mmap_error(err)) {
+// However, it is not clear that this loss of our reserved mapping
+// happens with large pages on Linux or that we cannot recover
+// from the loss. For now, we just issue a warning and we don't
+// call vm_exit_out_of_memory(). This issue is being tracked by
+// JBS-8007074.
+warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
+//    vm_exit_out_of_memory(size, OOM_MMAP_ERROR,
+//                          "committing reserved memory.");
 }
 // Fall through and try to use small pages
 }
-if (commit_memory(addr, size, exec)) {
+err = os::Linux::commit_memory_impl(addr, size, exec);
+if (err == 0) {
 realign_memory(addr, size, alignment_hint);
-return true;
+}
-}
+return err;
-return false;
+}
+bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
+bool exec) {
+return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
+}
+void os::pd_commit_memory_or_exit(char* addr, size_t size,
+size_t alignment_hint, bool exec,
+const char* mesg) {
+assert(mesg != NULL, "mesg must be specified");
+int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
+if (err != 0) {
+// the caller wants all commit errors to exit with the specified mesg:
+warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
+vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
+}
 }
 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
 if (UseHugeTLBFS && alignment_hint > (size_t)vm_page_size()) {
 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
 // the existing pages. However it won't work for SHM-based large pages that cannot be
 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
 // small pages on top of the SHM segment. This method always works for small pages, so we
 // allow that in any case.
 if (alignment_hint <= (size_t)os::vm_page_size() || !UseSHM) {
-commit_memory(addr, bytes, alignment_hint, false);
+commit_memory(addr, bytes, alignment_hint, !ExecMem);
 }
 }
 void os::numa_make_global(char *addr, size_t bytes) {
 Linux::numa_interleave_memory(addr, bytes);
 "growable stack in non-initial thread");
 if (stack_extent < (uintptr_t)addr)
 ::munmap((void*)stack_extent, (uintptr_t)addr - stack_extent);
 }
-return os::commit_memory(addr, size);
+return os::commit_memory(addr, size, !ExecMem);
 }
 // If this is a growable mapping, remove the guard pages entirely by
 // munmap()ping them.  If not, just call uncommit_memory(). This only
 // affects the main/initial thread, but guard against future OS changes
 bool result = false;
 void *p = mmap (NULL, page_size, PROT_READ|PROT_WRITE,
 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
 -1, 0);
-if (p != (void *) -1) {
+if (p != MAP_FAILED) {
 // We don't know if this really is a huge page or not.
 FILE *fp = fopen("/proc/self/maps", "r");
 if (fp) {
 while (!feof(fp)) {
 char chars[257];
 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
 #endif
 if (!UseMembar) {
 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
-guarantee( mem_serialize_page != NULL, "mmap Failed for memory serialize page");
+guarantee( mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
 os::set_memory_serialize_page( mem_serialize_page );
 #ifndef PRODUCT
 if(Verbose && PrintMiscellaneous)
 tty->print("[Memory Serialize  Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);

Mercurial > hg > graal-jvmci-8

comparison src/os/linux/vm/os_linux.cpp @ 10969:a837fa3d3f86