Fri, 17 Mar 2017 03:39:23 -0700
8049717: expose L1_data_cache_line_size for diagnostic/sanity checks
Summary: Add support for VM_Version::L1_data_cache_line_size().
Reviewed-by: dsimms, kvn, dholmes
1 /*
2 * Copyright (c) 1997, 2014, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
25 #include "precompiled.hpp"
26 #include "asm/macroAssembler.inline.hpp"
27 #include "memory/resourceArea.hpp"
28 #include "runtime/java.hpp"
29 #include "runtime/stubCodeGenerator.hpp"
30 #include "vm_version_sparc.hpp"
31 #ifdef TARGET_OS_FAMILY_linux
32 # include "os_linux.inline.hpp"
33 #endif
34 #ifdef TARGET_OS_FAMILY_solaris
35 # include "os_solaris.inline.hpp"
36 #endif
38 int VM_Version::_features = VM_Version::unknown_m;
39 const char* VM_Version::_features_str = "";
40 unsigned int VM_Version::_L2_data_cache_line_size = 0;
42 void VM_Version::initialize() {
44 assert(_features != VM_Version::unknown_m, "System pre-initialization is not complete.");
45 guarantee(VM_Version::has_v9(), "only SPARC v9 is supported");
47 PrefetchCopyIntervalInBytes = prefetch_copy_interval_in_bytes();
48 PrefetchScanIntervalInBytes = prefetch_scan_interval_in_bytes();
49 PrefetchFieldsAhead = prefetch_fields_ahead();
51 assert(0 <= AllocatePrefetchInstr && AllocatePrefetchInstr <= 1, "invalid value");
52 if( AllocatePrefetchInstr < 0 ) AllocatePrefetchInstr = 0;
53 if( AllocatePrefetchInstr > 1 ) AllocatePrefetchInstr = 0;
55 // Allocation prefetch settings
56 intx cache_line_size = prefetch_data_size();
57 if( cache_line_size > AllocatePrefetchStepSize )
58 AllocatePrefetchStepSize = cache_line_size;
60 assert(AllocatePrefetchLines > 0, "invalid value");
61 if( AllocatePrefetchLines < 1 ) // set valid value in product VM
62 AllocatePrefetchLines = 3;
63 assert(AllocateInstancePrefetchLines > 0, "invalid value");
64 if( AllocateInstancePrefetchLines < 1 ) // set valid value in product VM
65 AllocateInstancePrefetchLines = 1;
67 AllocatePrefetchDistance = allocate_prefetch_distance();
68 AllocatePrefetchStyle = allocate_prefetch_style();
70 assert((AllocatePrefetchDistance % AllocatePrefetchStepSize) == 0 &&
71 (AllocatePrefetchDistance > 0), "invalid value");
72 if ((AllocatePrefetchDistance % AllocatePrefetchStepSize) != 0 ||
73 (AllocatePrefetchDistance <= 0)) {
74 AllocatePrefetchDistance = AllocatePrefetchStepSize;
75 }
77 if (AllocatePrefetchStyle == 3 && (!has_blk_init() || cache_line_size <= 0)) {
78 warning("BIS instructions are not available on this CPU");
79 FLAG_SET_DEFAULT(AllocatePrefetchStyle, 1);
80 }
82 assert(ArraycopySrcPrefetchDistance < 4096, "invalid value");
83 if (ArraycopySrcPrefetchDistance >= 4096)
84 ArraycopySrcPrefetchDistance = 4064;
85 assert(ArraycopyDstPrefetchDistance < 4096, "invalid value");
86 if (ArraycopyDstPrefetchDistance >= 4096)
87 ArraycopyDstPrefetchDistance = 4064;
89 UseSSE = 0; // Only on x86 and x64
91 _supports_cx8 = has_v9();
92 _supports_atomic_getset4 = true; // swap instruction
94 // There are Fujitsu Sparc64 CPUs which support blk_init as well so
95 // we have to take this check out of the 'is_niagara()' block below.
96 if (has_blk_init()) {
97 // When using CMS or G1, we cannot use memset() in BOT updates
98 // because the sun4v/CMT version in libc_psr uses BIS which
99 // exposes "phantom zeros" to concurrent readers. See 6948537.
100 if (FLAG_IS_DEFAULT(UseMemSetInBOT) && (UseConcMarkSweepGC || UseG1GC)) {
101 FLAG_SET_DEFAULT(UseMemSetInBOT, false);
102 }
103 // Issue a stern warning if the user has explicitly set
104 // UseMemSetInBOT (it is known to cause issues), but allow
105 // use for experimentation and debugging.
106 if (UseConcMarkSweepGC || UseG1GC) {
107 if (UseMemSetInBOT) {
108 assert(!FLAG_IS_DEFAULT(UseMemSetInBOT), "Error");
109 warning("Experimental flag -XX:+UseMemSetInBOT is known to cause instability"
110 " on sun4v; please understand that you are using at your own risk!");
111 }
112 }
113 }
115 if (is_niagara()) {
116 // Indirect branch is the same cost as direct
117 if (FLAG_IS_DEFAULT(UseInlineCaches)) {
118 FLAG_SET_DEFAULT(UseInlineCaches, false);
119 }
120 // Align loops on a single instruction boundary.
121 if (FLAG_IS_DEFAULT(OptoLoopAlignment)) {
122 FLAG_SET_DEFAULT(OptoLoopAlignment, 4);
123 }
124 #ifdef _LP64
125 // 32-bit oops don't make sense for the 64-bit VM on sparc
126 // since the 32-bit VM has the same registers and smaller objects.
127 Universe::set_narrow_oop_shift(LogMinObjAlignmentInBytes);
128 Universe::set_narrow_klass_shift(LogKlassAlignmentInBytes);
129 #endif // _LP64
130 #ifdef COMPILER2
131 // Indirect branch is the same cost as direct
132 if (FLAG_IS_DEFAULT(UseJumpTables)) {
133 FLAG_SET_DEFAULT(UseJumpTables, true);
134 }
135 // Single-issue, so entry and loop tops are
136 // aligned on a single instruction boundary
137 if (FLAG_IS_DEFAULT(InteriorEntryAlignment)) {
138 FLAG_SET_DEFAULT(InteriorEntryAlignment, 4);
139 }
140 if (is_niagara_plus()) {
141 if (has_blk_init() && (cache_line_size > 0) && UseTLAB &&
142 FLAG_IS_DEFAULT(AllocatePrefetchInstr)) {
143 // Use BIS instruction for TLAB allocation prefetch.
144 FLAG_SET_ERGO(intx, AllocatePrefetchInstr, 1);
145 if (FLAG_IS_DEFAULT(AllocatePrefetchStyle)) {
146 FLAG_SET_ERGO(intx, AllocatePrefetchStyle, 3);
147 }
148 if (FLAG_IS_DEFAULT(AllocatePrefetchDistance)) {
149 // Use smaller prefetch distance with BIS
150 FLAG_SET_DEFAULT(AllocatePrefetchDistance, 64);
151 }
152 }
153 if (is_T4()) {
154 // Double number of prefetched cache lines on T4
155 // since L2 cache line size is smaller (32 bytes).
156 if (FLAG_IS_DEFAULT(AllocatePrefetchLines)) {
157 FLAG_SET_ERGO(intx, AllocatePrefetchLines, AllocatePrefetchLines*2);
158 }
159 if (FLAG_IS_DEFAULT(AllocateInstancePrefetchLines)) {
160 FLAG_SET_ERGO(intx, AllocateInstancePrefetchLines, AllocateInstancePrefetchLines*2);
161 }
162 }
163 if (AllocatePrefetchStyle != 3 && FLAG_IS_DEFAULT(AllocatePrefetchDistance)) {
164 // Use different prefetch distance without BIS
165 FLAG_SET_DEFAULT(AllocatePrefetchDistance, 256);
166 }
167 if (AllocatePrefetchInstr == 1) {
168 // Need a space at the end of TLAB for BIS since it
169 // will fault when accessing memory outside of heap.
171 // +1 for rounding up to next cache line, +1 to be safe
172 int lines = AllocatePrefetchLines + 2;
173 int step_size = AllocatePrefetchStepSize;
174 int distance = AllocatePrefetchDistance;
175 _reserve_for_allocation_prefetch = (distance + step_size*lines)/(int)HeapWordSize;
176 }
177 }
178 #endif
179 }
181 // Use hardware population count instruction if available.
182 if (has_hardware_popc()) {
183 if (FLAG_IS_DEFAULT(UsePopCountInstruction)) {
184 FLAG_SET_DEFAULT(UsePopCountInstruction, true);
185 }
186 } else if (UsePopCountInstruction) {
187 warning("POPC instruction is not available on this CPU");
188 FLAG_SET_DEFAULT(UsePopCountInstruction, false);
189 }
191 // T4 and newer Sparc cpus have new compare and branch instruction.
192 if (has_cbcond()) {
193 if (FLAG_IS_DEFAULT(UseCBCond)) {
194 FLAG_SET_DEFAULT(UseCBCond, true);
195 }
196 } else if (UseCBCond) {
197 warning("CBCOND instruction is not available on this CPU");
198 FLAG_SET_DEFAULT(UseCBCond, false);
199 }
201 assert(BlockZeroingLowLimit > 0, "invalid value");
202 if (has_block_zeroing() && cache_line_size > 0) {
203 if (FLAG_IS_DEFAULT(UseBlockZeroing)) {
204 FLAG_SET_DEFAULT(UseBlockZeroing, true);
205 }
206 } else if (UseBlockZeroing) {
207 warning("BIS zeroing instructions are not available on this CPU");
208 FLAG_SET_DEFAULT(UseBlockZeroing, false);
209 }
211 assert(BlockCopyLowLimit > 0, "invalid value");
212 if (has_block_zeroing() && cache_line_size > 0) { // has_blk_init() && is_T4(): core's local L2 cache
213 if (FLAG_IS_DEFAULT(UseBlockCopy)) {
214 FLAG_SET_DEFAULT(UseBlockCopy, true);
215 }
216 } else if (UseBlockCopy) {
217 warning("BIS instructions are not available or expensive on this CPU");
218 FLAG_SET_DEFAULT(UseBlockCopy, false);
219 }
221 #ifdef COMPILER2
222 // T4 and newer Sparc cpus have fast RDPC.
223 if (has_fast_rdpc() && FLAG_IS_DEFAULT(UseRDPCForConstantTableBase)) {
224 FLAG_SET_DEFAULT(UseRDPCForConstantTableBase, true);
225 }
227 // Currently not supported anywhere.
228 FLAG_SET_DEFAULT(UseFPUForSpilling, false);
230 MaxVectorSize = 8;
232 assert((InteriorEntryAlignment % relocInfo::addr_unit()) == 0, "alignment is not a multiple of NOP size");
233 #endif
235 assert((CodeEntryAlignment % relocInfo::addr_unit()) == 0, "alignment is not a multiple of NOP size");
236 assert((OptoLoopAlignment % relocInfo::addr_unit()) == 0, "alignment is not a multiple of NOP size");
238 char buf[512];
239 jio_snprintf(buf, sizeof(buf), "%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s",
240 (has_v9() ? ", v9" : (has_v8() ? ", v8" : "")),
241 (has_hardware_popc() ? ", popc" : ""),
242 (has_vis1() ? ", vis1" : ""),
243 (has_vis2() ? ", vis2" : ""),
244 (has_vis3() ? ", vis3" : ""),
245 (has_blk_init() ? ", blk_init" : ""),
246 (has_cbcond() ? ", cbcond" : ""),
247 (has_aes() ? ", aes" : ""),
248 (has_sha1() ? ", sha1" : ""),
249 (has_sha256() ? ", sha256" : ""),
250 (has_sha512() ? ", sha512" : ""),
251 (is_ultra3() ? ", ultra3" : ""),
252 (is_sun4v() ? ", sun4v" : ""),
253 (is_niagara_plus() ? ", niagara_plus" : (is_niagara() ? ", niagara" : "")),
254 (is_sparc64() ? ", sparc64" : ""),
255 (!has_hardware_mul32() ? ", no-mul32" : ""),
256 (!has_hardware_div32() ? ", no-div32" : ""),
257 (!has_hardware_fsmuld() ? ", no-fsmuld" : ""));
259 // buf is started with ", " or is empty
260 _features_str = strdup(strlen(buf) > 2 ? buf + 2 : buf);
262 // There are three 64-bit SPARC families that do not overlap, e.g.,
263 // both is_ultra3() and is_sparc64() cannot be true at the same time.
264 // Within these families, there can be more than one chip, e.g.,
265 // is_T4() and is_T7() machines are also is_niagara().
266 if (is_ultra3()) {
267 assert(_L1_data_cache_line_size == 0, "overlap with Ultra3 family");
268 // Ref: UltraSPARC III Cu Processor
269 _L1_data_cache_line_size = 64;
270 }
271 if (is_niagara()) {
272 assert(_L1_data_cache_line_size == 0, "overlap with niagara family");
273 // All Niagara's are sun4v's, but not all sun4v's are Niagaras, e.g.,
274 // Fujitsu SPARC64 is sun4v, but we don't want it in this block.
275 //
276 // Ref: UltraSPARC T1 Supplement to the UltraSPARC Architecture 2005
277 // Appendix F.1.3.1 Cacheable Accesses
278 // -> 16-byte L1 cache line size
279 //
280 // Ref: UltraSPARC T2: A Highly-Threaded, Power-Efficient, SPARC SOC
281 // Section III: SPARC Processor Core
282 // -> 16-byte L1 cache line size
283 //
284 // Ref: Oracle's SPARC T4-1, SPARC T4-2, SPARC T4-4, and SPARC T4-1B Server Architecture
285 // Section SPARC T4 Processor Cache Architecture
286 // -> 32-byte L1 cache line size (no longer see that info on this ref)
287 //
288 // XXX - still need a T7 reference here
289 //
290 if (is_T7()) { // T7 or newer
291 _L1_data_cache_line_size = 64;
292 } else if (is_T4()) { // T4 or newer (until T7)
293 _L1_data_cache_line_size = 32;
294 } else { // T1 or newer (until T4)
295 _L1_data_cache_line_size = 16;
296 }
297 }
298 if (is_sparc64()) {
299 guarantee(_L1_data_cache_line_size == 0, "overlap with SPARC64 family");
300 // Ref: Fujitsu SPARC64 VII Processor
301 // Section 4 Cache System
302 _L1_data_cache_line_size = 64;
303 }
305 // UseVIS is set to the smallest of what hardware supports and what
306 // the command line requires. I.e., you cannot set UseVIS to 3 on
307 // older UltraSparc which do not support it.
308 if (UseVIS > 3) UseVIS=3;
309 if (UseVIS < 0) UseVIS=0;
310 if (!has_vis3()) // Drop to 2 if no VIS3 support
311 UseVIS = MIN2((intx)2,UseVIS);
312 if (!has_vis2()) // Drop to 1 if no VIS2 support
313 UseVIS = MIN2((intx)1,UseVIS);
314 if (!has_vis1()) // Drop to 0 if no VIS1 support
315 UseVIS = 0;
317 // SPARC T4 and above should have support for AES instructions
318 if (has_aes()) {
319 if (UseVIS > 2) { // AES intrinsics use MOVxTOd/MOVdTOx which are VIS3
320 if (FLAG_IS_DEFAULT(UseAES)) {
321 FLAG_SET_DEFAULT(UseAES, true);
322 }
323 if (FLAG_IS_DEFAULT(UseAESIntrinsics)) {
324 FLAG_SET_DEFAULT(UseAESIntrinsics, true);
325 }
326 // we disable both the AES flags if either of them is disabled on the command line
327 if (!UseAES || !UseAESIntrinsics) {
328 FLAG_SET_DEFAULT(UseAES, false);
329 FLAG_SET_DEFAULT(UseAESIntrinsics, false);
330 }
331 } else {
332 if (UseAES || UseAESIntrinsics) {
333 warning("SPARC AES intrinsics require VIS3 instruction support. Intrinsics will be disabled.");
334 if (UseAES) {
335 FLAG_SET_DEFAULT(UseAES, false);
336 }
337 if (UseAESIntrinsics) {
338 FLAG_SET_DEFAULT(UseAESIntrinsics, false);
339 }
340 }
341 }
342 } else if (UseAES || UseAESIntrinsics) {
343 warning("AES instructions are not available on this CPU");
344 if (UseAES) {
345 FLAG_SET_DEFAULT(UseAES, false);
346 }
347 if (UseAESIntrinsics) {
348 FLAG_SET_DEFAULT(UseAESIntrinsics, false);
349 }
350 }
352 // SHA1, SHA256, and SHA512 instructions were added to SPARC T-series at different times
353 if (has_sha1() || has_sha256() || has_sha512()) {
354 if (UseVIS > 0) { // SHA intrinsics use VIS1 instructions
355 if (FLAG_IS_DEFAULT(UseSHA)) {
356 FLAG_SET_DEFAULT(UseSHA, true);
357 }
358 } else {
359 if (UseSHA) {
360 warning("SPARC SHA intrinsics require VIS1 instruction support. Intrinsics will be disabled.");
361 FLAG_SET_DEFAULT(UseSHA, false);
362 }
363 }
364 } else if (UseSHA) {
365 warning("SHA instructions are not available on this CPU");
366 FLAG_SET_DEFAULT(UseSHA, false);
367 }
369 if (!UseSHA) {
370 FLAG_SET_DEFAULT(UseSHA1Intrinsics, false);
371 FLAG_SET_DEFAULT(UseSHA256Intrinsics, false);
372 FLAG_SET_DEFAULT(UseSHA512Intrinsics, false);
373 } else {
374 if (has_sha1()) {
375 if (FLAG_IS_DEFAULT(UseSHA1Intrinsics)) {
376 FLAG_SET_DEFAULT(UseSHA1Intrinsics, true);
377 }
378 } else if (UseSHA1Intrinsics) {
379 warning("SHA1 instruction is not available on this CPU.");
380 FLAG_SET_DEFAULT(UseSHA1Intrinsics, false);
381 }
382 if (has_sha256()) {
383 if (FLAG_IS_DEFAULT(UseSHA256Intrinsics)) {
384 FLAG_SET_DEFAULT(UseSHA256Intrinsics, true);
385 }
386 } else if (UseSHA256Intrinsics) {
387 warning("SHA256 instruction (for SHA-224 and SHA-256) is not available on this CPU.");
388 FLAG_SET_DEFAULT(UseSHA256Intrinsics, false);
389 }
391 if (has_sha512()) {
392 if (FLAG_IS_DEFAULT(UseSHA512Intrinsics)) {
393 FLAG_SET_DEFAULT(UseSHA512Intrinsics, true);
394 }
395 } else if (UseSHA512Intrinsics) {
396 warning("SHA512 instruction (for SHA-384 and SHA-512) is not available on this CPU.");
397 FLAG_SET_DEFAULT(UseSHA512Intrinsics, false);
398 }
399 if (!(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics)) {
400 FLAG_SET_DEFAULT(UseSHA, false);
401 }
402 }
404 if (FLAG_IS_DEFAULT(ContendedPaddingWidth) &&
405 (cache_line_size > ContendedPaddingWidth))
406 ContendedPaddingWidth = cache_line_size;
408 #ifndef PRODUCT
409 if (PrintMiscellaneous && Verbose) {
410 tty->print_cr("L1 data cache line size: %u", L1_data_cache_line_size());
411 tty->print_cr("L2 data cache line size: %u", L2_data_cache_line_size());
412 tty->print("Allocation");
413 if (AllocatePrefetchStyle <= 0) {
414 tty->print_cr(": no prefetching");
415 } else {
416 tty->print(" prefetching: ");
417 if (AllocatePrefetchInstr == 0) {
418 tty->print("PREFETCH");
419 } else if (AllocatePrefetchInstr == 1) {
420 tty->print("BIS");
421 }
422 if (AllocatePrefetchLines > 1) {
423 tty->print_cr(" at distance %d, %d lines of %d bytes", (int) AllocatePrefetchDistance, (int) AllocatePrefetchLines, (int) AllocatePrefetchStepSize);
424 } else {
425 tty->print_cr(" at distance %d, one line of %d bytes", (int) AllocatePrefetchDistance, (int) AllocatePrefetchStepSize);
426 }
427 }
428 if (PrefetchCopyIntervalInBytes > 0) {
429 tty->print_cr("PrefetchCopyIntervalInBytes %d", (int) PrefetchCopyIntervalInBytes);
430 }
431 if (PrefetchScanIntervalInBytes > 0) {
432 tty->print_cr("PrefetchScanIntervalInBytes %d", (int) PrefetchScanIntervalInBytes);
433 }
434 if (PrefetchFieldsAhead > 0) {
435 tty->print_cr("PrefetchFieldsAhead %d", (int) PrefetchFieldsAhead);
436 }
437 if (ContendedPaddingWidth > 0) {
438 tty->print_cr("ContendedPaddingWidth %d", (int) ContendedPaddingWidth);
439 }
440 }
441 #endif // PRODUCT
442 }
444 void VM_Version::print_features() {
445 tty->print_cr("Version:%s", cpu_features());
446 }
448 int VM_Version::determine_features() {
449 if (UseV8InstrsOnly) {
450 NOT_PRODUCT(if (PrintMiscellaneous && Verbose) tty->print_cr("Version is Forced-V8");)
451 return generic_v8_m;
452 }
454 int features = platform_features(unknown_m); // platform_features() is os_arch specific
456 if (features == unknown_m) {
457 features = generic_v9_m;
458 warning("Cannot recognize SPARC version. Default to V9");
459 }
461 assert(is_T_family(features) == is_niagara(features), "Niagara should be T series");
462 if (UseNiagaraInstrs) { // Force code generation for Niagara
463 if (is_T_family(features)) {
464 // Happy to accomodate...
465 } else {
466 NOT_PRODUCT(if (PrintMiscellaneous && Verbose) tty->print_cr("Version is Forced-Niagara");)
467 features |= T_family_m;
468 }
469 } else {
470 if (is_T_family(features) && !FLAG_IS_DEFAULT(UseNiagaraInstrs)) {
471 NOT_PRODUCT(if (PrintMiscellaneous && Verbose) tty->print_cr("Version is Forced-Not-Niagara");)
472 features &= ~(T_family_m | T1_model_m);
473 } else {
474 // Happy to accomodate...
475 }
476 }
478 return features;
479 }
481 static int saved_features = 0;
483 void VM_Version::allow_all() {
484 saved_features = _features;
485 _features = all_features_m;
486 }
488 void VM_Version::revert() {
489 _features = saved_features;
490 }
492 unsigned int VM_Version::calc_parallel_worker_threads() {
493 unsigned int result;
494 if (is_M_series()) {
495 // for now, use same gc thread calculation for M-series as for niagara-plus
496 // in future, we may want to tweak parameters for nof_parallel_worker_thread
497 result = nof_parallel_worker_threads(5, 16, 8);
498 } else if (is_niagara_plus()) {
499 result = nof_parallel_worker_threads(5, 16, 8);
500 } else {
501 result = nof_parallel_worker_threads(5, 8, 8);
502 }
503 return result;
504 }