8278411: Implement UseHeavyMonitors consistently, s390 port

Reviewed-by: mdoerr, stuefe, lucy
This commit is contained in:
Amit Kumar 2023-05-16 15:25:19 +00:00 committed by Martin Doerr
parent 316bc79e0e
commit 41ee125a0f
5 changed files with 99 additions and 71 deletions
src/hotspot
test/jdk/java/util/concurrent/ConcurrentHashMap

@ -86,7 +86,6 @@ void C1_MacroAssembler::verified_entry(bool breakAtEntry) {
void C1_MacroAssembler::lock_object(Register hdr, Register obj, Register disp_hdr, Label& slow_case) {
const int hdr_offset = oopDesc::mark_offset_in_bytes();
assert_different_registers(hdr, obj, disp_hdr);
NearLabel done;
verify_oop(obj, FILE_AND_LINE);
@ -102,40 +101,47 @@ void C1_MacroAssembler::lock_object(Register hdr, Register obj, Register disp_hd
z_btrue(slow_case);
}
// and mark it as unlocked.
z_oill(hdr, markWord::unlocked_value);
// Save unlocked object header into the displaced header location on the stack.
z_stg(hdr, Address(disp_hdr, (intptr_t)0));
// Test if object header is still the same (i.e. unlocked), and if so, store the
// displaced header address in the object header. If it is not the same, get the
// object header instead.
z_csg(hdr, disp_hdr, hdr_offset, obj);
// If the object header was the same, we're done.
branch_optimized(Assembler::bcondEqual, done);
// If the object header was not the same, it is now in the hdr register.
// => Test if it is a stack pointer into the same stack (recursive locking), i.e.:
//
// 1) (hdr & markWord::lock_mask_in_place) == 0
// 2) rsp <= hdr
// 3) hdr <= rsp + page_size
//
// These 3 tests can be done by evaluating the following expression:
//
// (hdr - Z_SP) & (~(page_size-1) | markWord::lock_mask_in_place)
//
// assuming both the stack pointer and page_size have their least
// significant 2 bits cleared and page_size is a power of 2
z_sgr(hdr, Z_SP);
assert(LockingMode != LM_MONITOR, "LM_MONITOR is already handled, by emit_lock()");
load_const_optimized(Z_R0_scratch, (~(os::vm_page_size()-1) | markWord::lock_mask_in_place));
z_ngr(hdr, Z_R0_scratch); // AND sets CC (result eq/ne 0).
// For recursive locking, the result is zero. => Save it in the displaced header
// location (null in the displaced hdr location indicates recursive locking).
z_stg(hdr, Address(disp_hdr, (intptr_t)0));
// Otherwise we don't care about the result and handle locking via runtime call.
branch_optimized(Assembler::bcondNotZero, slow_case);
// done
bind(done);
if (LockingMode == LM_LIGHTWEIGHT) {
Unimplemented();
} else if (LockingMode == LM_LEGACY) {
NearLabel done;
// and mark it as unlocked.
z_oill(hdr, markWord::unlocked_value);
// Save unlocked object header into the displaced header location on the stack.
z_stg(hdr, Address(disp_hdr, (intptr_t) 0));
// Test if object header is still the same (i.e. unlocked), and if so, store the
// displaced header address in the object header. If it is not the same, get the
// object header instead.
z_csg(hdr, disp_hdr, hdr_offset, obj);
// If the object header was the same, we're done.
branch_optimized(Assembler::bcondEqual, done);
// If the object header was not the same, it is now in the hdr register.
// => Test if it is a stack pointer into the same stack (recursive locking), i.e.:
//
// 1) (hdr & markWord::lock_mask_in_place) == 0
// 2) rsp <= hdr
// 3) hdr <= rsp + page_size
//
// These 3 tests can be done by evaluating the following expression:
//
// (hdr - Z_SP) & (~(page_size-1) | markWord::lock_mask_in_place)
//
// assuming both the stack pointer and page_size have their least
// significant 2 bits cleared and page_size is a power of 2
z_sgr(hdr, Z_SP);
load_const_optimized(Z_R0_scratch, (~(os::vm_page_size() - 1) | markWord::lock_mask_in_place));
z_ngr(hdr, Z_R0_scratch); // AND sets CC (result eq/ne 0).
// For recursive locking, the result is zero. => Save it in the displaced header
// location (null in the displaced hdr location indicates recursive locking).
z_stg(hdr, Address(disp_hdr, (intptr_t) 0));
// Otherwise we don't care about the result and handle locking via runtime call.
branch_optimized(Assembler::bcondNotZero, slow_case);
// done
bind(done);
}
}
void C1_MacroAssembler::unlock_object(Register hdr, Register obj, Register disp_hdr, Label& slow_case) {
@ -144,21 +150,29 @@ void C1_MacroAssembler::unlock_object(Register hdr, Register obj, Register disp_
assert_different_registers(hdr, obj, disp_hdr);
NearLabel done;
// Load displaced header.
z_ltg(hdr, Address(disp_hdr, (intptr_t)0));
// If the loaded hdr is null we had recursive locking, and we are done.
z_bre(done);
if (LockingMode != LM_LIGHTWEIGHT) {
// Load displaced header.
z_ltg(hdr, Address(disp_hdr, (intptr_t) 0));
// If the loaded hdr is null we had recursive locking, and we are done.
z_bre(done);
}
// Load object.
z_lg(obj, Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes()));
verify_oop(obj, FILE_AND_LINE);
// Test if object header is pointing to the displaced header, and if so, restore
// the displaced header in the object. If the object header is not pointing to
// the displaced header, get the object header instead.
z_csg(disp_hdr, hdr, hdr_offset, obj);
// If the object header was not pointing to the displaced header,
// we do unlocking via runtime call.
branch_optimized(Assembler::bcondNotEqual, slow_case);
// done
if (LockingMode == LM_LIGHTWEIGHT) {
Unimplemented();
} else {
// Test if object header is pointing to the displaced header, and if so, restore
// the displaced header in the object. If the object header is not pointing to
// the displaced header, get the object header instead.
z_csg(disp_hdr, hdr, hdr_offset, obj);
// If the object header was not pointing to the displaced header,
// we do unlocking via runtime call.
branch_optimized(Assembler::bcondNotEqual, slow_case);
// done
}
bind(done);
}

@ -3171,22 +3171,28 @@ void MacroAssembler::compiler_fast_lock_object(Register oop, Register box, Regis
z_nill(temp, markWord::monitor_value);
z_brne(object_has_monitor);
// Set mark to markWord | markWord::unlocked_value.
z_oill(displacedHeader, markWord::unlocked_value);
if (LockingMode != LM_MONITOR) {
// Set mark to markWord | markWord::unlocked_value.
z_oill(displacedHeader, markWord::unlocked_value);
// Load Compare Value application register.
// Load Compare Value application register.
// Initialize the box (must happen before we update the object mark).
z_stg(displacedHeader, BasicLock::displaced_header_offset_in_bytes(), box);
// Initialize the box (must happen before we update the object mark).
z_stg(displacedHeader, BasicLock::displaced_header_offset_in_bytes(), box);
// Memory Fence (in cmpxchgd)
// Compare object markWord with mark and if equal exchange scratch1 with object markWord.
// Memory Fence (in cmpxchgd)
// Compare object markWord with mark and if equal exchange scratch1 with object markWord.
// If the compare-and-swap succeeded, then we found an unlocked object and we
// have now locked it.
z_csg(displacedHeader, box, 0, oop);
assert(currentHeader==displacedHeader, "must be same register"); // Identified two registers from z/Architecture.
z_bre(done);
// If the compare-and-swap succeeded, then we found an unlocked object and we
// have now locked it.
z_csg(displacedHeader, box, 0, oop);
assert(currentHeader == displacedHeader, "must be same register"); // Identified two registers from z/Architecture.
z_bre(done);
} else {
// Set NE to indicate 'failure' -> take slow-path
z_ltgr(oop, oop);
z_bru(done);
}
// We did not see an unlocked object so try the fast recursive case.
@ -3238,10 +3244,12 @@ void MacroAssembler::compiler_fast_unlock_object(Register oop, Register box, Reg
BLOCK_COMMENT("compiler_fast_unlock_object {");
// Find the lock address and load the displaced header from the stack.
// if the displaced header is zero, we have a recursive unlock.
load_and_test_long(displacedHeader, Address(box, BasicLock::displaced_header_offset_in_bytes()));
z_bre(done);
if (LockingMode != LM_MONITOR) {
// Find the lock address and load the displaced header from the stack.
// if the displaced header is zero, we have a recursive unlock.
load_and_test_long(displacedHeader, Address(box, BasicLock::displaced_header_offset_in_bytes()));
z_bre(done);
}
// Handle existing monitor.
// The object has an existing monitor iff (mark & monitor_value) != 0.
@ -3250,12 +3258,18 @@ void MacroAssembler::compiler_fast_unlock_object(Register oop, Register box, Reg
z_nill(currentHeader, markWord::monitor_value);
z_brne(object_has_monitor);
// Check if it is still a light weight lock, this is true if we see
// the stack address of the basicLock in the markWord of the object
// copy box to currentHeader such that csg does not kill it.
z_lgr(currentHeader, box);
z_csg(currentHeader, displacedHeader, 0, oop);
z_bru(done); // Csg sets CR as desired.
if (LockingMode != LM_MONITOR) {
// Check if it is still a light weight lock, this is true if we see
// the stack address of the basicLock in the markWord of the object
// copy box to currentHeader such that csg does not kill it.
z_lgr(currentHeader, box);
z_csg(currentHeader, displacedHeader, 0, oop);
z_bru(done); // Csg sets CR as desired.
} else {
// Set NE to indicate 'failure' -> take slow-path
z_ltgr(oop, oop);
z_bru(done);
}
// Handle existing monitor.
bind(object_has_monitor);

@ -1921,7 +1921,7 @@ bool Arguments::check_vm_args_consistency() {
FLAG_SET_CMDLINE(LockingMode, LM_MONITOR);
}
#if !defined(X86) && !defined(AARCH64) && !defined(PPC64) && !defined(RISCV64)
#if !defined(X86) && !defined(AARCH64) && !defined(PPC64) && !defined(RISCV64) && !defined(S390)
if (LockingMode == LM_MONITOR) {
jio_fprintf(defaultStream::error_stream(),
"LockingMode == 0 (LM_MONITOR) is not fully implemented on this architecture");

@ -486,7 +486,7 @@ void ObjectSynchronizer::handle_sync_on_value_based_class(Handle obj, JavaThread
}
static bool useHeavyMonitors() {
#if defined(X86) || defined(AARCH64) || defined(PPC64) || defined(RISCV64)
#if defined(X86) || defined(AARCH64) || defined(PPC64) || defined(RISCV64) || defined(S390)
return LockingMode == LM_MONITOR;
#else
return false;

@ -48,7 +48,7 @@
/*
* @test
* @summary Exercise multithreaded maps, using only heavy monitors.
* @requires os.arch=="x86" | os.arch=="i386" | os.arch=="amd64" | os.arch=="x86_64" | os.arch=="aarch64" | os.arch == "ppc64" | os.arch == "ppc64le" | os.arch == "riscv64"
* @requires os.arch=="x86" | os.arch=="i386" | os.arch=="amd64" | os.arch=="x86_64" | os.arch=="aarch64" | os.arch == "ppc64" | os.arch == "ppc64le" | os.arch == "riscv64" | os.arch == "s390x"
* @requires vm.debug
* @library /test/lib
* @run main/othervm/timeout=1600 -XX:+UseHeavyMonitors -XX:+VerifyHeavyMonitors MapLoops