1 /*
   2  * Copyright (c) 1998, 2007, 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  */
  24 
  25 #include "incls/_precompiled.incl"
  26 #include "incls/_threadLS_solaris_x86.cpp.incl"
  27 
  28 #ifdef AMD64
  29 extern "C" Thread*  fs_load(ptrdiff_t tlsOffset);
  30 extern "C" intptr_t fs_thread();
  31 #else
  32 // From solaris_i486.s
  33 extern "C" Thread*  gs_load(ptrdiff_t tlsOffset);
  34 extern "C" intptr_t gs_thread();
  35 #endif // AMD64
  36 
  37 // tlsMode encoding:
  38 //
  39 // pd_tlsAccessUndefined : uninitialized
  40 // pd_tlsAccessSlow      : not available
  41 // pd_tlsAccessIndirect  :
  42 //   old-style indirect access - present in "T1" libthread.
  43 //   use thr_slot_sync_allocate() to attempt to allocate a slot.
  44 // pd_tlsAccessDirect    :
  45 //   new-style direct access - present in late-model "T2" libthread.
  46 //   Allocate the offset (slot) via _thr_slot_offset() or by
  47 //   defining an IE- or LE-mode TLS/TSD slot in the launcher and then passing
  48 //   that offset into libjvm.so.
  49 //   See http://sac.eng/Archives/CaseLog/arc/PSARC/2003/159/.
  50 //
  51 // Note that we have a capability gap - some early model T2 forms
  52 // (e.g., unpatched S9) have neither _thr_slot_sync_allocate() nor
  53 // _thr_slot_offset().  In that case we revert to the usual
  54 // thr_getspecific accessor.
  55 //
  56 
  57 static ThreadLocalStorage::pd_tlsAccessMode tlsMode = ThreadLocalStorage::pd_tlsAccessUndefined ;
  58 static ptrdiff_t tlsOffset = 0 ;
  59 static thread_key_t tlsKey ;
  60 
  61 typedef int (*TSSA_Entry) (ptrdiff_t *, int, int) ;
  62 typedef ptrdiff_t (*TSO_Entry) (int) ;
  63 
  64 ThreadLocalStorage::pd_tlsAccessMode ThreadLocalStorage::pd_getTlsAccessMode ()
  65 {
  66    guarantee (tlsMode != pd_tlsAccessUndefined, "tlsMode not set") ;
  67    return tlsMode ;
  68 }
  69 
  70 ptrdiff_t ThreadLocalStorage::pd_getTlsOffset () {
  71    guarantee (tlsMode != pd_tlsAccessUndefined, "tlsMode not set") ;
  72    return tlsOffset ;
  73 }
  74 
  75 // TODO: Consider the following improvements:
  76 //
  77 // 1.   Convert from thr_*specific* to pthread_*specific*.
  78 //      The pthread_ forms are slightly faster.  Also, the
  79 //      pthread_ forms have a pthread_key_delete() API which
  80 //      would aid in clean JVM shutdown and the eventual goal
  81 //      of permitting a JVM to reinstantiate itself withing a process.
  82 //
  83 // 2.   See ThreadLocalStorage::init().  We end up allocating
  84 //      two TLS keys during VM startup.  That's benign, but we could collapse
  85 //      down to one key without too much trouble.
  86 //
  87 // 3.   MacroAssembler::get_thread() currently emits calls to thr_getspecific().
  88 //      Modify get_thread() to call Thread::current() instead.
  89 //
  90 // 4.   Thread::current() currently uses a cache keyed by %gs:[0].
  91 //      (The JVM has PSARC permission to use %g7/%gs:[0]
  92 //      as an opaque temporally unique thread identifier).
  93 //      For C++ access to a thread's reflexive "self" pointer we
  94 //      should consider using one of the following:
  95 //      a. a radix tree keyed by %esp - as in EVM.
  96 //         This requires two loads (the 2nd dependent on the 1st), but
  97 //         is easily inlined and doesn't require a "miss" slow path.
  98 //      b. a fast TLS/TSD slot allocated by _thr_slot_offset
  99 //         or _thr_slot_sync_allocate.
 100 //
 101 // 5.   'generate_code_for_get_thread' is a misnomer.
 102 //      We should change it to something more general like
 103 //      pd_ThreadSelf_Init(), for instance.
 104 //
 105 
 106 static void AllocateTLSOffset ()
 107 {
 108    int rslt ;
 109    TSSA_Entry tssa ;
 110    TSO_Entry  tso ;
 111    ptrdiff_t off ;
 112 
 113    guarantee (tlsMode == ThreadLocalStorage::pd_tlsAccessUndefined, "tlsMode not set") ;
 114    tlsMode = ThreadLocalStorage::pd_tlsAccessSlow ;
 115    tlsOffset = 0 ;
 116 #ifndef AMD64
 117 
 118    tssa = (TSSA_Entry) dlsym (RTLD_DEFAULT, "thr_slot_sync_allocate") ;
 119    if (tssa != NULL) {
 120         off = -1 ;
 121         rslt = (*tssa)(&off, NULL, NULL) ;                // (off,dtor,darg)
 122         if (off != -1) {
 123            tlsOffset = off ;
 124            tlsMode = ThreadLocalStorage::pd_tlsAccessIndirect ;
 125            return ;
 126         }
 127     }
 128 
 129     rslt = thr_keycreate (&tlsKey, NULL) ;
 130     if (rslt != 0) {
 131         tlsMode = ThreadLocalStorage::pd_tlsAccessSlow ;   // revert to slow mode
 132         return ;
 133     }
 134 
 135     tso = (TSO_Entry) dlsym (RTLD_DEFAULT, "_thr_slot_offset") ;
 136     if (tso != NULL) {
 137         off = (*tso)(tlsKey) ;
 138         if (off >= 0) {
 139            tlsOffset = off ;
 140            tlsMode = ThreadLocalStorage::pd_tlsAccessDirect ;
 141            return ;
 142         }
 143     }
 144 
 145     // Failure: Too bad ... we've allocated a TLS slot we don't need and there's
 146     // no provision in the ABI for returning the slot.
 147     //
 148     // If we didn't find a slot then then:
 149     // 1. We might be on liblwp.
 150     // 2. We might be on T2 libthread, but all "fast" slots are already
 151     //    consumed
 152     // 3. We might be on T1, and all TSD (thr_slot_sync_allocate) slots are
 153     //    consumed.
 154     // 4. We might be on T2 libthread, but it's be re-architected
 155     //    so that fast slots are no longer g7-relative.
 156     //
 157 
 158     tlsMode = ThreadLocalStorage::pd_tlsAccessSlow ;
 159     return ;
 160 #endif // AMD64
 161 }
 162 
 163 void ThreadLocalStorage::generate_code_for_get_thread() {
 164     AllocateTLSOffset() ;
 165 }
 166 
 167 void ThreadLocalStorage::set_thread_in_slot(Thread *thread) {
 168   guarantee (tlsMode != pd_tlsAccessUndefined, "tlsMode not set") ;
 169   if (tlsMode == pd_tlsAccessIndirect) {
 170 #ifdef AMD64
 171         intptr_t tbase = fs_thread();
 172 #else
 173         intptr_t tbase = gs_thread();
 174 #endif // AMD64
 175         *((Thread**) (tbase + tlsOffset)) = thread ;
 176   } else
 177   if (tlsMode == pd_tlsAccessDirect) {
 178         thr_setspecific (tlsKey, (void *) thread) ;
 179         // set with thr_setspecific and then readback with gs_load to validate.
 180 #ifdef AMD64
 181         guarantee (thread == fs_load(tlsOffset), "tls readback failure") ;
 182 #else
 183         guarantee (thread == gs_load(tlsOffset), "tls readback failure") ;
 184 #endif // AMD64
 185   }
 186 }
 187 
 188 
 189 extern "C" Thread* get_thread() {
 190   return ThreadLocalStorage::thread();
 191 }