-/* context_sysv - context switching with ucontextes from System V */
+/* context_sysv - context switching with ucontexts from System V */
-/* Copyright (c) 2009, 2010. The SimGrid Team.
+/* Copyright (c) 2009-2014. The SimGrid Team.
* All rights reserved. */
- /* This program is free software; you can redistribute it and/or modify it
- * under the terms of the license (GNU LGPL) which comes with this package. */
+/* This program is free software; you can redistribute it and/or modify it
+ * under the terms of the license (GNU LGPL) which comes with this package. */
#include <stdarg.h>
#include "xbt/parmap.h"
#include "smx_private.h"
-#include "gras_config.h"
+#include "internal_config.h"
#include "context_sysv_config.h" /* loads context system definitions */
+#include "mc/mc.h"
#ifdef _XBT_WIN32
-# include <win32_ucontext.h> /* context relative declarations */
+# include <xbt/win32_ucontext.h> /* context relative declarations */
#else
# include <ucontext.h> /* context relative declarations */
#endif
-#ifdef HAVE_VALGRIND_VALGRIND_H
-# include <valgrind/valgrind.h>
-#endif /* HAVE_VALGRIND_VALGRIND_H */
-
XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(simix_context);
typedef struct s_smx_ctx_sysv {
s_smx_ctx_base_t super; /* Fields of super implementation */
ucontext_t uc; /* the ucontext that executes the code */
-#ifdef HAVE_VALGRIND_VALGRIND_H
- unsigned int valgrind_stack_id; /* the valgrind stack id */
-#endif
- char stack[0]; /* the thread stack (must remain the last element of the structure) */
+ char *stack; /* the thread stack */
} s_smx_ctx_sysv_t, *smx_ctx_sysv_t;
#ifdef CONTEXT_THREADS
static xbt_parmap_t sysv_parmap;
-static ucontext_t* sysv_workers_stacks; /* space to save the worker's stack in each thread */
+static smx_ctx_sysv_t* sysv_workers_context; /* space to save the worker's context in each thread */
static unsigned long sysv_threads_working; /* number of threads that have started their work */
static xbt_os_thread_key_t sysv_worker_id_key; /* thread-specific storage for the thread id */
#endif
static smx_ctx_sysv_t sysv_maestro_context;
static int smx_ctx_sysv_factory_finalize(smx_context_factory_t *factory);
-static smx_context_t
-smx_ctx_sysv_create_context_sized(size_t structure_size,
- xbt_main_func_t code, int argc,
- char **argv,
- void_pfn_smxprocess_t cleanup_func,
- void *data);
static void smx_ctx_sysv_free(smx_context_t context);
static smx_context_t
smx_ctx_sysv_create_context(xbt_main_func_t code, int argc, char **argv,
- void_pfn_smxprocess_t cleanup_func, void* data);
+ void_pfn_smxprocess_t cleanup_func, smx_process_t process);
static void smx_ctx_sysv_wrapper(int count, ...);
static void smx_ctx_sysv_resume_parallel(smx_process_t first_process);
static void smx_ctx_sysv_runall_parallel(void);
-/* This is a bit paranoid about SIZEOF_VOIDP not being a multiple of SIZEOF_INT,
- * but it doesn't harm. */
-#define CTX_ADDR_LEN (SIZEOF_VOIDP / SIZEOF_INT + !!(SIZEOF_VOIDP % SIZEOF_INT))
-union u_ctx_addr {
- void *addr;
- int intv[CTX_ADDR_LEN];
-};
-#if (CTX_ADDR_LEN == 1)
-# define CTX_ADDR_SPLIT(u) (u).intv[0]
-#elif (CTX_ADDR_LEN == 2)
-# define CTX_ADDR_SPLIT(u) (u).intv[0], (u).intv[1]
-#else
-# error Your architecture is not supported yet
-#endif
+/* This is a bit paranoid about sizeof(smx_ctx_sysv_t) not being a multiple of
+ * sizeof(int), but it doesn't harm. */
+#define CTX_ADDR_LEN \
+ (sizeof(smx_ctx_sysv_t) / sizeof(int) + \
+ !!(sizeof(smx_ctx_sysv_t) % sizeof(int)))
void SIMIX_ctx_sysv_factory_init(smx_context_factory_t *factory)
{
(*factory)->name = "smx_sysv_context_factory";
if (SIMIX_context_is_parallel()) {
-#ifdef CONTEXT_THREADS /* To use parallel ucontexts a thread pool is needed */
+#ifdef CONTEXT_THREADS /* To use parallel ucontexts a thread pool is needed */
int nthreads = SIMIX_context_get_nthreads();
sysv_parmap = xbt_parmap_new(nthreads, SIMIX_context_get_parallel_mode());
- sysv_workers_stacks = xbt_new(ucontext_t, nthreads);
+ sysv_workers_context = xbt_new(smx_ctx_sysv_t, nthreads);
+ sysv_maestro_context = NULL;
xbt_os_thread_key_create(&sysv_worker_id_key);
(*factory)->stop = smx_ctx_sysv_stop_parallel;
(*factory)->suspend = smx_ctx_sysv_suspend_parallel;
#ifdef CONTEXT_THREADS
if (sysv_parmap)
xbt_parmap_destroy(sysv_parmap);
- xbt_free(sysv_workers_stacks);
+ xbt_free(sysv_workers_context);
#endif
return smx_ctx_base_factory_finalize(factory);
}
static smx_context_t
-smx_ctx_sysv_create_context_sized(size_t size, xbt_main_func_t code,
- int argc, char **argv,
- void_pfn_smxprocess_t cleanup_func,
- void *data)
+smx_ctx_sysv_create_context(xbt_main_func_t code, int argc, char **argv,
+ void_pfn_smxprocess_t cleanup_func,
+ smx_process_t process)
{
- union u_ctx_addr ctx_addr;
+ int ctx_addr[CTX_ADDR_LEN];
smx_ctx_sysv_t context =
- (smx_ctx_sysv_t) smx_ctx_base_factory_create_context_sized(size,
- code,
- argc,
- argv,
- cleanup_func,
- data);
+ (smx_ctx_sysv_t) smx_ctx_base_factory_create_context_sized(
+ sizeof(s_smx_ctx_sysv_t),
+ code,
+ argc,
+ argv,
+ cleanup_func,
+ process);
/* if the user provided a function for the process then use it,
otherwise it is the context for maestro */
if (code) {
+ context->stack = SIMIX_context_stack_new();
getcontext(&(context->uc));
context->uc.uc_link = NULL;
context->uc.uc_stack.ss_sp =
- pth_skaddr_makecontext(context->stack, smx_context_stack_size);
+ pth_skaddr_makecontext(context->stack, smx_context_usable_stack_size);
context->uc.uc_stack.ss_size =
- pth_sksize_makecontext(context->stack, smx_context_stack_size);
-
-#ifdef HAVE_VALGRIND_VALGRIND_H
- context->valgrind_stack_id =
- VALGRIND_STACK_REGISTER(context->uc.uc_stack.ss_sp,
- ((char *) context->uc.uc_stack.ss_sp) +
- context->uc.uc_stack.ss_size);
-#endif /* HAVE_VALGRIND_VALGRIND_H */
- ctx_addr.addr = context;
- makecontext(&context->uc, (void (*)())smx_ctx_sysv_wrapper,
- CTX_ADDR_LEN, CTX_ADDR_SPLIT(ctx_addr));
+ pth_sksize_makecontext(context->stack, smx_context_usable_stack_size);
+
+ memcpy(ctx_addr, &context, sizeof(smx_ctx_sysv_t));
+ switch (CTX_ADDR_LEN) {
+ case 1:
+ makecontext(&context->uc, (void (*)())smx_ctx_sysv_wrapper,
+ 1, ctx_addr[0]);
+ break;
+ case 2:
+ makecontext(&context->uc, (void (*)())smx_ctx_sysv_wrapper,
+ 2, ctx_addr[0], ctx_addr[1]);
+ break;
+ default:
+ xbt_die("Ucontexts are not supported on this arch yet (addr len = %zu/%zu = %zu)",
+ sizeof(smx_ctx_sysv_t), sizeof(int), CTX_ADDR_LEN);
+ }
} else {
- sysv_maestro_context = context;
+ if (process != NULL && sysv_maestro_context == NULL)
+ sysv_maestro_context = context;
}
- return (smx_context_t) context;
-}
-
-static smx_context_t
-smx_ctx_sysv_create_context(xbt_main_func_t code, int argc, char **argv,
- void_pfn_smxprocess_t cleanup_func,
- void *data)
-{
-
- return smx_ctx_sysv_create_context_sized(sizeof(s_smx_ctx_sysv_t) + smx_context_stack_size,
- code, argc, argv, cleanup_func,
- data);
+#ifdef HAVE_MC
+ if (MC_is_active() && code) {
+ MC_register_stack_area(context->stack, ((smx_context_t)context)->process,
+ &(context->uc), smx_context_usable_stack_size);
+ }
+#endif
+ return (smx_context_t) context;
}
static void smx_ctx_sysv_free(smx_context_t context)
{
if (context) {
-
-#ifdef HAVE_VALGRIND_VALGRIND_H
- VALGRIND_STACK_DEREGISTER(((smx_ctx_sysv_t)
- context)->valgrind_stack_id);
-#endif /* HAVE_VALGRIND_VALGRIND_H */
-
+ SIMIX_context_stack_delete(((smx_ctx_sysv_t)context)->stack);
}
smx_ctx_base_free(context);
}
static void smx_ctx_sysv_wrapper(int first, ...)
{
- union u_ctx_addr ctx_addr;
+ int ctx_addr[CTX_ADDR_LEN];
smx_ctx_sysv_t context;
- ctx_addr.intv[0] = first;
+ ctx_addr[0] = first;
if (CTX_ADDR_LEN > 1) {
va_list ap;
int i;
va_start(ap, first);
for (i = 1; i < CTX_ADDR_LEN; i++)
- ctx_addr.intv[i] = va_arg(ap, int);
+ ctx_addr[i] = va_arg(ap, int);
va_end(ap);
}
- context = ctx_addr.addr;
+ memcpy(&context, ctx_addr, sizeof(smx_ctx_sysv_t));
(context->super.code) (context->super.argc, context->super.argv);
simix_global->context_factory->stop((smx_context_t) context);
static void smx_ctx_sysv_runall_serial(void)
{
- if (!xbt_dynar_is_empty(simix_global->process_to_run)) {
- smx_process_t first_process =
- xbt_dynar_get_as(simix_global->process_to_run, 0, smx_process_t);
- sysv_process_index = 1;
+ smx_process_t first_process =
+ xbt_dynar_get_as(simix_global->process_to_run, 0, smx_process_t);
+ sysv_process_index = 1;
- /* execute the first process */
- smx_ctx_sysv_resume_serial(first_process);
- }
+ /* execute the first process */
+ smx_ctx_sysv_resume_serial(first_process);
}
static void smx_ctx_sysv_stop_parallel(smx_context_t context)
smx_ctx_sysv_suspend_parallel(context);
}
-static void smx_ctx_sysv_suspend_parallel(smx_context_t context)
-{
+/* This function is called by maestro at the beginning of a scheduling round to get all working threads executing some stuff
+ * It is much easier to understand what happens if you see the working threads as bodies that swap their soul for the
+ * ones of the simulated processes that must run.
+ */
+static void smx_ctx_sysv_runall_parallel(void) {
+#ifdef CONTEXT_THREADS
+ sysv_threads_working = 0;
+ // parmap_apply ensures that every working thread get an index in the process_to_run array (through an atomic fetch_and_add),
+ // and runs the smx_ctx_sysv_resume_parallel function on that index
+ xbt_parmap_apply(sysv_parmap, (void_f_pvoid_t) smx_ctx_sysv_resume_parallel,
+ simix_global->process_to_run);
+#else
+ xbt_die("You asked for a parallel execution, but you don't have any threads.");
+#endif
+}
+
+/* This function is in charge of running one particular simulated process on the current thread */
+static void smx_ctx_sysv_resume_parallel(smx_process_t simulated_process_to_run) {
+#ifdef CONTEXT_THREADS
+ unsigned long worker_id = __sync_fetch_and_add(&sysv_threads_working, 1); // what is my containing body?
+ xbt_os_thread_set_specific(sysv_worker_id_key, (void*) worker_id); // Store the number of my containing body in os-thread-specific area
+ smx_ctx_sysv_t worker_context = (smx_ctx_sysv_t)SIMIX_context_self(); // get my current soul
+ sysv_workers_context[worker_id] = worker_context; // write down that this soul is hosted in that body (for now)
+ ucontext_t* worker_stack = &worker_context->uc; // retrieves the system-level info that fuels this soul
+
+ smx_context_t context = simulated_process_to_run->context; // That's the first soul that I should become
+ SIMIX_context_set_current(context); // write in simix that I switched my soul
+ swapcontext(worker_stack, &((smx_ctx_sysv_t) context)->uc); // actually do that using the relevant syscall
+ // No body runs that soul anymore at this point. Instead the current body took the soul of simulated process
+ // The simulated process wakes back after the call to "SIMIX_context_suspend(self->context);" within smx_process.c::SIMIX_process_yield()
+
+ // From now on, the simulated processes will change their soul with the next soul to execute (in suspend_parallel, below).
+ // When nobody is to be executed in this scheduling round, the last simulated process will take back the initial soul of the current working thread
+#endif
+}
+
+/* This function is called when a simulated process wants to yield back to the maestro in a blocking simcall.
+ * This naturally occurs within SIMIX_context_suspend(self->context), called from SIMIX_process_yield()
+ * Actually, it does not really yield back to maestro, but into the next process that must be executed.
+ * If no one is to be executed, then it yields to the initial soul that was in this working thread (that was saved in resume_parallel).
+ */
+static void smx_ctx_sysv_suspend_parallel(smx_context_t context) {
#ifdef CONTEXT_THREADS
/* determine the next context */
- smx_process_t next_work = xbt_parmap_next(sysv_parmap);
+ smx_process_t next_work = xbt_parmap_next(sysv_parmap); // get the next soul to embody now
smx_context_t next_context;
- ucontext_t* next_stack;
+ ucontext_t* next_stack; // will contain the next soul to run, either simulated or initial minion's one
- if (next_work != NULL) {
- /* there is a next process to resume */
+ if (next_work != NULL) { // there is a next soul to embody (ie, a next process to resume)
XBT_DEBUG("Run next process");
next_context = next_work->context;
- next_stack = &((smx_ctx_sysv_t) next_context)->uc;
}
else {
/* all processes were run, go to the barrier */
XBT_DEBUG("No more processes to run");
- next_context = (smx_context_t) sysv_maestro_context;
- unsigned long worker_id =
+ unsigned long worker_id = // Get back the identity of my body that was stored when starting the scheduling round
(unsigned long) xbt_os_thread_get_specific(sysv_worker_id_key);
- next_stack = &sysv_workers_stacks[worker_id];
+ next_context = (smx_context_t)sysv_workers_context[worker_id]; // deduce the initial soul of that body
+ // When given that soul, the body will wait for the next scheduling round
}
- SIMIX_context_set_current(next_context);
- swapcontext(&((smx_ctx_sysv_t) context)->uc, next_stack);
-#endif
-}
+ next_stack = &((smx_ctx_sysv_t)next_context)->uc;
-static void smx_ctx_sysv_resume_parallel(smx_process_t first_process)
-{
-#ifdef CONTEXT_THREADS
- unsigned long worker_id = __sync_fetch_and_add(&sysv_threads_working, 1);
- xbt_os_thread_set_specific(sysv_worker_id_key, (void*) worker_id);
- ucontext_t* worker_stack = &sysv_workers_stacks[worker_id];
-
- smx_context_t context = first_process->context;
- SIMIX_context_set_current(context);
- swapcontext(worker_stack, &((smx_ctx_sysv_t) context)->uc);
-#endif
-}
-
-static void smx_ctx_sysv_runall_parallel(void)
-{
-#ifdef CONTEXT_THREADS
- sysv_threads_working = 0;
- xbt_parmap_apply(sysv_parmap, (void_f_pvoid_t) smx_ctx_sysv_resume_parallel,
- simix_global->process_to_run);
+ SIMIX_context_set_current(next_context);
+ swapcontext(&((smx_ctx_sysv_t) context)->uc, next_stack); // get that next soul
#endif
}