fix raw contextes for supernovae and for parallel execution

[simgrid.git] / src / simix / smx_context_raw.c

diff --git a/src/simix/smx_context_raw.c b/src/simix/smx_context_raw.c
index 51d4c0d..b91dfd6 100644 (file)
--- a/src/simix/smx_context_raw.c
+++ b/src/simix/smx_context_raw.c
@@ -33,38 +33,43 @@ typedef struct s_smx_ctx_raw {
   s_smx_ctx_base_t super;       /* Fields of super implementation */
   char *malloced_stack; /* malloced area containing the stack */
   raw_stack_t stack_top; /* pointer to stack top (within previous area) */
   s_smx_ctx_base_t super;       /* Fields of super implementation */
   char *malloced_stack; /* malloced area containing the stack */
   raw_stack_t stack_top; /* pointer to stack top (within previous area) */

+  raw_stack_t old_stack_top; /* to whom I should return the control */

 #ifdef HAVE_VALGRIND_VALGRIND_H
   unsigned int valgrind_stack_id;       /* the valgrind stack id */
 #endif
 } s_smx_ctx_raw_t, *smx_ctx_raw_t;
 
 #ifdef HAVE_VALGRIND_VALGRIND_H
   unsigned int valgrind_stack_id;       /* the valgrind stack id */
 #endif
 } s_smx_ctx_raw_t, *smx_ctx_raw_t;
 

-smx_ctx_raw_t maestro_context;
+smx_ctx_raw_t maestro_raw_context;

 
 extern raw_stack_t raw_makecontext(char* malloced_stack, int stack_size,
                                    rawctx_entry_point_t entry_point, void* arg);
 
 extern raw_stack_t raw_makecontext(char* malloced_stack, int stack_size,
                                    rawctx_entry_point_t entry_point, void* arg);

-extern void raw_swapcontext(raw_stack_t* old, raw_stack_t* new);
+extern void raw_swapcontext(raw_stack_t* old, raw_stack_t new);

 
 #ifdef PROCESSOR_i686
 __asm__ (
 
 #ifdef PROCESSOR_i686
 __asm__ (

-".globl raw_makecontext\n"
-"raw_makecontext:\n"
-"   movl 4(%esp),%eax\n"  /* stack */
-"   addl 8(%esp),%eax\n"  /* size  */
-"   movl 12(%esp),%ecx\n" /* func  */
-"   movl 16(%esp),%edx\n" /* arg   */
-"   movl %edx, -4(%eax)\n"
-"   movl $0,   -8(%eax)\n"
-"   movl %ecx,-12(%eax)\n"
-"   movl $0,  -16(%eax)\n"
-"   movl $0,  -20(%eax)\n"
-"   movl $0,  -24(%eax)\n"
-"   movl $0,  -28(%eax)\n"
-"   subl $28,%eax\n"
-"   ret\n"
+   ".text\n"
+   ".globl raw_makecontext\n"
+   ".type raw_makecontext,@function\n"
+   "raw_makecontext:\n"
+   "   movl 4(%esp),%eax\n"   /* stack */
+   "   addl 8(%esp),%eax\n"   /* size  */
+   "   movl 12(%esp),%ecx\n"  /* func  */
+   "   movl 16(%esp),%edx\n"  /* arg   */
+   "   movl %edx, -4(%eax)\n"
+   "   movl $0,   -8(%eax)\n" /* @return for func */
+   "   movl %ecx,-12(%eax)\n"
+   "   movl $0,  -16(%eax)\n" /* ebp */
+   "   movl $0,  -20(%eax)\n" /* ebx */
+   "   movl $0,  -24(%eax)\n" /* esi */
+   "   movl $0,  -28(%eax)\n" /* edi */
+   "   subl $28,%eax\n"
+   "   retl\n"

 );
 
 __asm__ (
 );
 
 __asm__ (

+   ".text\n"

    ".globl raw_swapcontext\n"
    ".globl raw_swapcontext\n"

+   ".type raw_swapcontext,@function\n"

    "raw_swapcontext:\n"
    "   movl 4(%esp),%eax\n" /* old */
    "   movl 8(%esp),%edx\n" /* new */
    "raw_swapcontext:\n"
    "   movl 4(%esp),%eax\n" /* old */
    "   movl 8(%esp),%edx\n" /* new */


@@ -73,12 +78,71 @@ __asm__ (
    "   pushl %esi\n"
    "   pushl %edi\n"
    "   movl %esp,(%eax)\n"
    "   pushl %esi\n"
    "   pushl %edi\n"
    "   movl %esp,(%eax)\n"

-   "   movl (%edx),%esp\n"
+   "   movl %edx,%esp\n"

    "   popl %edi\n"
    "   popl %esi\n"
    "   popl %ebx\n"
    "   popl %ebp\n"
    "   popl %edi\n"
    "   popl %esi\n"
    "   popl %ebx\n"
    "   popl %ebp\n"

-   "   ret\n"
+   "   retl\n"
+);
+#elif PROCESSOR_x86_64
+__asm__ (
+   ".text\n"
+   ".globl raw_makecontext\n"
+   ".type raw_makecontext,@function\n"
+   "raw_makecontext:\n" /* Calling convention sets the arguments in rdi, rsi, rdx and rcx, respectively */
+   "   movq %rdi,%rax\n"      /* stack */
+   "   addq %rsi,%rax\n"      /* size  */
+   "   movq $0,   -8(%rax)\n" /* @return for func */
+   "   movq %rdx,-16(%rax)\n" /* func */
+   "   movq %rcx,-24(%rax)\n" /* arg/rdi */
+   "   movq $0,  -32(%rax)\n" /* rsi */
+   "   movq $0,  -40(%rax)\n" /* rdx */
+   "   movq $0,  -48(%rax)\n" /* rcx */
+   "   movq $0,  -56(%rax)\n" /* r8  */
+   "   movq $0,  -64(%rax)\n" /* r9  */
+   "   movq $0,  -72(%rax)\n" /* rbp */
+   "   movq $0,  -80(%rax)\n" /* rbx */
+   "   movq $0,  -88(%rax)\n" /* r12 */
+   "   movq $0,  -96(%rax)\n" /* r13 */
+   "   movq $0, -104(%rax)\n" /* r14 */
+   "   movq $0, -112(%rax)\n" /* r15 */
+   "   subq $112,%rax\n"
+   "   retq\n"
+);
+
+__asm__ (
+   ".text\n"
+   ".globl raw_swapcontext\n"
+   ".type raw_swapcontext,@function\n"
+   "raw_swapcontext:\n" /* Calling convention sets the arguments in rdi and rsi, respectively */
+   "   pushq %rdi\n"
+   "   pushq %rsi\n"
+   "   pushq %rdx\n"
+   "   pushq %rcx\n"
+   "   pushq %r8\n"
+   "   pushq %r9\n"
+   "   pushq %rbp\n"
+   "   pushq %rbx\n"
+   "   pushq %r12\n"
+   "   pushq %r13\n"
+   "   pushq %r14\n"
+   "   pushq %r15\n"
+   "   movq %rsp,(%rdi)\n" /* old */
+   "   movq %rsi,%rsp\n" /* new */
+   "   popq %r15\n"
+   "   popq %r14\n"
+   "   popq %r13\n"
+   "   popq %r12\n"
+   "   popq %rbx\n"
+   "   popq %rbp\n"
+   "   popq %r9\n"
+   "   popq %r8\n"
+   "   popq %rcx\n"
+   "   popq %rdx\n"
+   "   popq %rsi\n"
+   "   popq %rdi\n"
+   "   retq\n"

 );
 #else
 raw_stack_t raw_makecontext(char* malloced_stack, int stack_size,
 );
 #else
 raw_stack_t raw_makecontext(char* malloced_stack, int stack_size,


@@ -86,7 +150,7 @@ raw_stack_t raw_makecontext(char* malloced_stack, int stack_size,
    THROW_UNIMPLEMENTED;
 }
 
    THROW_UNIMPLEMENTED;
 }
 

-void raw_swapcontext(raw_stack_t* old, raw_stack_t* new) {
+void raw_swapcontext(raw_stack_t* old, raw_stack_t new) {

    THROW_UNIMPLEMENTED;
 }
 
    THROW_UNIMPLEMENTED;
 }
 


@@ -136,7 +200,7 @@ smx_ctx_raw_create_context_sized(size_t size, xbt_main_func_t code,
 #endif                          /* HAVE_VALGRIND_VALGRIND_H */
 
      }else{
 #endif                          /* HAVE_VALGRIND_VALGRIND_H */
 
      }else{

-       maestro_context = context;
+       maestro_raw_context = context;

      }
 
      return (smx_context_t) context;
      }
 
      return (smx_context_t) context;


@@ -171,8 +235,10 @@ static void smx_ctx_raw_free(smx_context_t context)
 
 static void smx_ctx_raw_suspend(smx_context_t context)
 {
 
 static void smx_ctx_raw_suspend(smx_context_t context)
 {

-  smx_current_context = (smx_context_t)maestro_context;
-  raw_swapcontext(&((smx_ctx_raw_t) context)->stack_top, &maestro_context->stack_top);
+  smx_current_context = (smx_context_t)maestro_raw_context;
+  raw_swapcontext(
+      &((smx_ctx_raw_t) context)->stack_top,
+      ((smx_ctx_raw_t) context)->old_stack_top);

 }
 
 static void smx_ctx_raw_stop(smx_context_t context)
 }
 
 static void smx_ctx_raw_stop(smx_context_t context)


@@ -191,7 +257,9 @@ static void smx_ctx_raw_wrapper(smx_ctx_raw_t context)
 static void smx_ctx_raw_resume(smx_context_t context)
 {
   smx_current_context = context; 
 static void smx_ctx_raw_resume(smx_context_t context)
 {
   smx_current_context = context; 

-  raw_swapcontext(&maestro_context->stack_top, &((smx_ctx_raw_t) context)->stack_top);
+  raw_swapcontext(
+      &((smx_ctx_raw_t) context)->old_stack_top,
+      ((smx_ctx_raw_t) context)->stack_top);

 }
 
 static void smx_ctx_raw_runall(xbt_swag_t processes)
 }
 
 static void smx_ctx_raw_runall(xbt_swag_t processes)


@@ -205,7 +273,10 @@ static void smx_ctx_raw_runall(xbt_swag_t processes)
 static void smx_ctx_raw_resume_parallel(smx_context_t context)
 {
   xbt_os_thread_set_extra_data(context);
 static void smx_ctx_raw_resume_parallel(smx_context_t context)
 {
   xbt_os_thread_set_extra_data(context);

-  raw_swapcontext(&maestro_context->stack_top, &((smx_ctx_raw_t) context)->stack_top);
+  raw_swapcontext(
+      &((smx_ctx_raw_t) context)->old_stack_top,
+      ((smx_ctx_raw_t) context)->stack_top);
+  xbt_os_thread_set_extra_data(NULL);

 }
 
 static void smx_ctx_raw_runall_parallel(xbt_swag_t processes)
 }
 
 static void smx_ctx_raw_runall_parallel(xbt_swag_t processes)


@@ -220,11 +291,12 @@ static void smx_ctx_raw_runall_parallel(xbt_swag_t processes)
 static smx_context_t smx_ctx_raw_self_parallel(void)
 {
   smx_context_t self_context = (smx_context_t) xbt_os_thread_get_extra_data();
 static smx_context_t smx_ctx_raw_self_parallel(void)
 {
   smx_context_t self_context = (smx_context_t) xbt_os_thread_get_extra_data();

-  return self_context ? self_context : (smx_context_t) maestro_context;
+  return self_context ? self_context : (smx_context_t) maestro_raw_context;

 }
 
 void SIMIX_ctx_raw_factory_init(smx_context_factory_t *factory)
 {
 }
 
 void SIMIX_ctx_raw_factory_init(smx_context_factory_t *factory)
 {

+  INFO0("Using raw contexts. Because the glibc is just not good enough for us.");

   smx_ctx_base_factory_init(factory);
 
   (*factory)->finalize  = smx_ctx_raw_factory_finalize;
   smx_ctx_base_factory_init(factory);
 
   (*factory)->finalize  = smx_ctx_raw_factory_finalize;