util/coroutine-ucontext.c

   1 /*
   2  * ucontext coroutine initialization code
   3  *
   4  * Copyright (C) 2006  Anthony Liguori <anthony@codemonkey.ws>
   5  * Copyright (C) 2011  Kevin Wolf <kwolf@redhat.com>
   6  *
   7  * This library is free software; you can redistribute it and/or
   8  * modify it under the terms of the GNU Lesser General Public
   9  * License as published by the Free Software Foundation; either
  10  * version 2.0 of the License, or (at your option) any later version.
  11  *
  12  * This library is distributed in the hope that it will be useful,
  13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * Lesser General Public License for more details.
  16  *
  17  * You should have received a copy of the GNU Lesser General Public
  18  * License along with this library; if not, see <http://www.gnu.org/licenses/>.
  19  */
  20
  21 /* XXX Is there a nicer way to disable glibc's stack check for longjmp? */
  22 #undef _FORTIFY_SOURCE
  23 #define _FORTIFY_SOURCE 0
  24
  25 #include "qemu/osdep.h"
  26 #include <ucontext.h>
  27 #include "qemu/coroutine_int.h"
  28 #include "qemu/coroutine-tls.h"
  29
  30 #ifdef CONFIG_VALGRIND_H
  31 #include <valgrind/valgrind.h>
  32 #endif
  33
  34 #ifdef QEMU_SANITIZE_ADDRESS
  35 #ifdef CONFIG_ASAN_IFACE_FIBER
  36 #define CONFIG_ASAN 1
  37 #include <sanitizer/asan_interface.h>
  38 #endif
  39 #endif
  40
  41 #ifdef CONFIG_TSAN
  42 #include <sanitizer/tsan_interface.h>
  43 #endif
  44
  45 typedef struct {
  46     Coroutine base;
  47     void *stack;
  48     size_t stack_size;
  49 #ifdef CONFIG_SAFESTACK
  50     /* Need an unsafe stack for each coroutine */
  51     void *unsafe_stack;
  52     size_t unsafe_stack_size;
  53 #endif
  54     sigjmp_buf env;
  55
  56 #ifdef CONFIG_TSAN
  57     void *tsan_co_fiber;
  58     void *tsan_caller_fiber;
  59 #endif
  60
  61 #ifdef CONFIG_VALGRIND_H
  62     unsigned int valgrind_stack_id;
  63 #endif
  64
  65 } CoroutineUContext;
  66
  67 /**
  68  * Per-thread coroutine bookkeeping
  69  */
  70 QEMU_DEFINE_STATIC_CO_TLS(Coroutine *, current);
  71 QEMU_DEFINE_STATIC_CO_TLS(CoroutineUContext, leader);
  72
  73 /*
  74  * va_args to makecontext() must be type 'int', so passing
  75  * the pointer we need may require several int args. This
  76  * union is a quick hack to let us do that
  77  */
  78 union cc_arg {
  79     void *p;
  80     int i[2];
  81 };
  82
  83 /*
  84  * QEMU_ALWAYS_INLINE only does so if __OPTIMIZE__, so we cannot use it.
  85  * always_inline is required to avoid TSan runtime fatal errors.
  86  */
  87 static inline __attribute__((always_inline))
  88 void on_new_fiber(CoroutineUContext *co)
  89 {
  90 #ifdef CONFIG_TSAN
  91     co->tsan_co_fiber = __tsan_create_fiber(0); /* flags: sync on switch */
  92     co->tsan_caller_fiber = __tsan_get_current_fiber();
  93 #endif
  94 }
  95
  96 /* always_inline is required to avoid TSan runtime fatal errors. */
  97 static inline __attribute__((always_inline))
  98 void finish_switch_fiber(void *fake_stack_save)
  99 {
 100 #ifdef CONFIG_ASAN
 101     CoroutineUContext *leaderp = get_ptr_leader();
 102     const void *bottom_old;
 103     size_t size_old;
 104
 105     __sanitizer_finish_switch_fiber(fake_stack_save, &bottom_old, &size_old);
 106
 107     if (!leaderp->stack) {
 108         leaderp->stack = (void *)bottom_old;
 109         leaderp->stack_size = size_old;
 110     }
 111 #endif
 112 #ifdef CONFIG_TSAN
 113     if (fake_stack_save) {
 114         __tsan_release(fake_stack_save);
 115         __tsan_switch_to_fiber(fake_stack_save, 0);  /* 0=synchronize */
 116     }
 117 #endif
 118 }
 119
 120 /* always_inline is required to avoid TSan runtime fatal errors. */
 121 static inline __attribute__((always_inline))
 122 void start_switch_fiber_asan(void **fake_stack_save,
 123                              const void *bottom, size_t size)
 124 {
 125 #ifdef CONFIG_ASAN
 126     __sanitizer_start_switch_fiber(fake_stack_save, bottom, size);
 127 #endif
 128 }
 129
 130 /* always_inline is required to avoid TSan runtime fatal errors. */
 131 static inline __attribute__((always_inline))
 132 void start_switch_fiber_tsan(void **fake_stack_save,
 133                              CoroutineUContext *co,
 134                              bool caller)
 135 {
 136 #ifdef CONFIG_TSAN
 137     void *new_fiber = caller ?
 138                       co->tsan_caller_fiber :
 139                       co->tsan_co_fiber;
 140     void *curr_fiber = __tsan_get_current_fiber();
 141     __tsan_acquire(curr_fiber);
 142
 143     *fake_stack_save = curr_fiber;
 144     __tsan_switch_to_fiber(new_fiber, 0);  /* 0=synchronize */
 145 #endif
 146 }
 147
 148 static void coroutine_trampoline(int i0, int i1)
 149 {
 150     union cc_arg arg;
 151     CoroutineUContext *self;
 152     Coroutine *co;
 153     void *fake_stack_save = NULL;
 154
 155     finish_switch_fiber(NULL);
 156
 157     arg.i[0] = i0;
 158     arg.i[1] = i1;
 159     self = arg.p;
 160     co = &self->base;
 161
 162     /* Initialize longjmp environment and switch back the caller */
 163     if (!sigsetjmp(self->env, 0)) {
 164         CoroutineUContext *leaderp = get_ptr_leader();
 165
 166         start_switch_fiber_asan(&fake_stack_save,
 167                                 leaderp->stack, leaderp->stack_size);
 168         start_switch_fiber_tsan(&fake_stack_save, self, true); /* true=caller */
 169         siglongjmp(*(sigjmp_buf *)co->entry_arg, 1);
 170     }
 171
 172     finish_switch_fiber(fake_stack_save);
 173
 174     while (true) {
 175         co->entry(co->entry_arg);
 176         qemu_coroutine_switch(co, co->caller, COROUTINE_TERMINATE);
 177     }
 178 }
 179
 180 Coroutine *qemu_coroutine_new(void)
 181 {
 182     CoroutineUContext *co;
 183     ucontext_t old_uc, uc;
 184     sigjmp_buf old_env;
 185     union cc_arg arg = {0};
 186     void *fake_stack_save = NULL;
 187
 188     /* The ucontext functions preserve signal masks which incurs a
 189      * system call overhead.  sigsetjmp(buf, 0)/siglongjmp() does not
 190      * preserve signal masks but only works on the current stack.
 191      * Since we need a way to create and switch to a new stack, use
 192      * the ucontext functions for that but sigsetjmp()/siglongjmp() for
 193      * everything else.
 194      */
 195
 196     if (getcontext(&uc) == -1) {
 197         abort();
 198     }
 199
 200     co = g_malloc0(sizeof(*co));
 201     co->stack_size = COROUTINE_STACK_SIZE;
 202     co->stack = qemu_alloc_stack(&co->stack_size);
 203 #ifdef CONFIG_SAFESTACK
 204     co->unsafe_stack_size = COROUTINE_STACK_SIZE;
 205     co->unsafe_stack = qemu_alloc_stack(&co->unsafe_stack_size);
 206 #endif
 207     co->base.entry_arg = &old_env; /* stash away our jmp_buf */
 208
 209     uc.uc_link = &old_uc;
 210     uc.uc_stack.ss_sp = co->stack;
 211     uc.uc_stack.ss_size = co->stack_size;
 212     uc.uc_stack.ss_flags = 0;
 213
 214 #ifdef CONFIG_VALGRIND_H
 215     co->valgrind_stack_id =
 216         VALGRIND_STACK_REGISTER(co->stack, co->stack + co->stack_size);
 217 #endif
 218
 219     arg.p = co;
 220
 221     on_new_fiber(co);
 222     makecontext(&uc, (void (*)(void))coroutine_trampoline,
 223                 2, arg.i[0], arg.i[1]);
 224
 225     /* swapcontext() in, siglongjmp() back out */
 226     if (!sigsetjmp(old_env, 0)) {
 227         start_switch_fiber_asan(&fake_stack_save, co->stack, co->stack_size);
 228         start_switch_fiber_tsan(&fake_stack_save,
 229                                 co, false); /* false=not caller */
 230
 231 #ifdef CONFIG_SAFESTACK
 232         /*
 233          * Before we swap the context, set the new unsafe stack
 234          * The unsafe stack grows just like the normal stack, so start from
 235          * the last usable location of the memory area.
 236          * NOTE: we don't have to re-set the usp afterwards because we are
 237          * coming back to this context through a siglongjmp.
 238          * The compiler already wrapped the corresponding sigsetjmp call with
 239          * code that saves the usp on the (safe) stack before the call, and
 240          * restores it right after (which is where we return with siglongjmp).
 241          */
 242         void *usp = co->unsafe_stack + co->unsafe_stack_size;
 243         __safestack_unsafe_stack_ptr = usp;
 244 #endif
 245
 246         swapcontext(&old_uc, &uc);
 247     }
 248
 249     finish_switch_fiber(fake_stack_save);
 250
 251     return &co->base;
 252 }
 253
 254 #ifdef CONFIG_VALGRIND_H
 255 /* Work around an unused variable in the valgrind.h macro... */
 256 #if !defined(__clang__)
 257 #pragma GCC diagnostic push
 258 #pragma GCC diagnostic ignored "-Wunused-but-set-variable"
 259 #endif
 260 static inline void valgrind_stack_deregister(CoroutineUContext *co)
 261 {
 262     VALGRIND_STACK_DEREGISTER(co->valgrind_stack_id);
 263 }
 264 #if !defined(__clang__)
 265 #pragma GCC diagnostic pop
 266 #endif
 267 #endif
 268
 269 #if defined(CONFIG_ASAN) && defined(CONFIG_COROUTINE_POOL)
 270 static void coroutine_fn terminate_asan(void *opaque)
 271 {
 272     CoroutineUContext *to = DO_UPCAST(CoroutineUContext, base, opaque);
 273
 274     set_current(opaque);
 275     start_switch_fiber_asan(NULL, to->stack, to->stack_size);
 276     G_STATIC_ASSERT(!IS_ENABLED(CONFIG_TSAN));
 277     siglongjmp(to->env, COROUTINE_ENTER);
 278 }
 279 #endif
 280
 281 void qemu_coroutine_delete(Coroutine *co_)
 282 {
 283     CoroutineUContext *co = DO_UPCAST(CoroutineUContext, base, co_);
 284
 285 #if defined(CONFIG_ASAN) && defined(CONFIG_COROUTINE_POOL)
 286     co_->entry_arg = qemu_coroutine_self();
 287     co_->entry = terminate_asan;
 288     qemu_coroutine_switch(co_->entry_arg, co_, COROUTINE_ENTER);
 289 #endif
 290
 291 #ifdef CONFIG_VALGRIND_H
 292     valgrind_stack_deregister(co);
 293 #endif
 294
 295     qemu_free_stack(co->stack, co->stack_size);
 296 #ifdef CONFIG_SAFESTACK
 297     qemu_free_stack(co->unsafe_stack, co->unsafe_stack_size);
 298 #endif
 299     g_free(co);
 300 }
 301
 302 /* This function is marked noinline to prevent GCC from inlining it
 303  * into coroutine_trampoline(). If we allow it to do that then it
 304  * hoists the code to get the address of the TLS variable "current"
 305  * out of the while() loop. This is an invalid transformation because
 306  * the sigsetjmp() call may be called when running thread A but
 307  * return in thread B, and so we might be in a different thread
 308  * context each time round the loop.
 309  */
 310 CoroutineAction __attribute__((noinline))
 311 qemu_coroutine_switch(Coroutine *from_, Coroutine *to_,
 312                       CoroutineAction action)
 313 {
 314     CoroutineUContext *from = DO_UPCAST(CoroutineUContext, base, from_);
 315     CoroutineUContext *to = DO_UPCAST(CoroutineUContext, base, to_);
 316     int ret;
 317     void *fake_stack_save = NULL;
 318
 319     set_current(to_);
 320
 321     ret = sigsetjmp(from->env, 0);
 322     if (ret == 0) {
 323         start_switch_fiber_asan(IS_ENABLED(CONFIG_COROUTINE_POOL) ||
 324                                 action != COROUTINE_TERMINATE ?
 325                                     &fake_stack_save : NULL,
 326                                 to->stack, to->stack_size);
 327         start_switch_fiber_tsan(&fake_stack_save,
 328                                 to, false); /* false=not caller */
 329         siglongjmp(to->env, action);
 330     }
 331
 332     finish_switch_fiber(fake_stack_save);
 333
 334     return ret;
 335 }
 336
 337 Coroutine *qemu_coroutine_self(void)
 338 {
 339     Coroutine *self = get_current();
 340     CoroutineUContext *leaderp = get_ptr_leader();
 341
 342     if (!self) {
 343         self = &leaderp->base;
 344         set_current(self);
 345     }
 346 #ifdef CONFIG_TSAN
 347     if (!leaderp->tsan_co_fiber) {
 348         leaderp->tsan_co_fiber = __tsan_get_current_fiber();
 349     }
 350 #endif
 351     return self;
 352 }
 353
 354 bool qemu_in_coroutine(void)
 355 {
 356     Coroutine *self = get_current();
 357
 358     return self && self->caller;
 359 }