--- /dev/null
+/* \r
+ * This file is part of the Palacios Virtual Machine Monitor developed\r
+ * by the V3VEE Project with funding from the United States National \r
+ * Science Foundation and the Department of Energy. \r
+ *\r
+ * The V3VEE Project is a joint project between Northwestern University\r
+ * and the University of New Mexico. You can find out more at \r
+ * http://www.v3vee.org\r
+ *\r
+ * Copyright (c) 2012, The V3VEE Project <http://www.v3vee.org> \r
+ * All rights reserved.\r
+ *\r
+ * Author: Oscar Mondragon <omondrag@cs.unm.edu>\r
+ * Patrick G. Bridges <bridges@cs.unm.edu>\r
+ *\r
+ * This is free software. You are permitted to use,\r
+ * redistribute, and modify it as specified in the file "V3VEE_LICENSE".\r
+ */\r
+\r
+\r
+#include <palacios/vmm.h>\r
+#include <palacios/vmm_time.h>\r
+#include <palacios/vm_guest.h>\r
+#include <palacios/vmm_hashtable.h>\r
+#include <palacios/vmm_config.h>\r
+#include <palacios/vmm_extensions.h>\r
+#include <palacios/vmm_edf_sched.h>\r
+\r
+\r
+\r
+#ifndef V3_CONFIG_DEBUG_EDF_SCHED\r
+#undef PrintDebug\r
+#define PrintDebug(fmt, args...)\r
+#endif\r
+\r
+/* Overview \r
+ *\r
+ * EDF Scheduling\r
+ *\r
+ * The EDF scheduler uses a dynamic calculated priority as scheduling criteria to choose\r
+ * what thread will be scheduled.That priority is calculated according with the relative \r
+ * deadline of the threads that are ready to run in the runqueue. This runqueue is a per-logical\r
+ * core data structure used to keep the runnable virtual cores (threads) allocated to that \r
+ * logical core.The threads with less time before its deadline will receive better priorities. \r
+ * The runqueue is sorted each time that a vCPU becomes runnable. At that time the vCPU is \r
+ * enqueue and a new scheduling decision is taken. Each time a vCPU is scheduled, the parameter\r
+ * slice used time is set to zero and the current deadline is calculated using its period. Once\r
+ * the vCPU uses the logical core for slice seconds, that vCPU sleeps until its next scheduling \r
+ * period (when is re-inserted in the runqueue) and yields the CPU to allow the scheduling \r
+ * of the vCPU with best priority in the runqueue. \r
+ */\r
+\r
+// Default configuration values for the EDF Scheduler\r
+// time parameters in microseconds \r
+\r
+#define MAX_PERIOD 1000000000\r
+#define MIN_PERIOD 50000\r
+#define MAX_SLICE 1000000000\r
+#define MIN_SLICE 10000\r
+#define CPU_PERCENT 100\r
+\r
+\r
+/*\r
+ * init_edf_config: Initialize scheduler configuration\r
+ */\r
+\r
+static void \r
+init_edf_config(struct vm_edf_sched_config *edf_config){\r
+\r
+ edf_config->min_slice = MIN_SLICE;\r
+ edf_config->max_slice = MAX_SLICE;\r
+ edf_config->min_period = MIN_PERIOD;\r
+ edf_config->max_period = MAX_PERIOD;\r
+ edf_config->cpu_percent = CPU_PERCENT;\r
+}\r
+\r
+\r
+/*\r
+ * edf_sched_init: Initialize the run queue\r
+ */\r
+\r
+int \r
+edf_sched_init(struct v3_vm_info *vm){\r
+\r
+ PrintDebug(vm, VCORE_NONE,"EDF Sched. Initializing vm %s\n", vm->name);\r
+\r
+ struct vm_sched_state *sched_state = &vm->sched; \r
+ sched_state->priv_data = V3_Malloc( vm->avail_cores * sizeof(struct vm_edf_rq));\r
+\r
+ if (!sched_state->priv_data) {\r
+ PrintError(vm, VCORE_NONE,"Cannot allocate in priv_data in edf_sched_init\n");\r
+ return -1;\r
+ }\r
+\r
+ int lcore = 0;\r
+ \r
+ PrintDebug(vm, VCORE_NONE,"EDF Sched. edf_sched_init. Available cores %d\n", vm->avail_cores);\r
+\r
+ for(lcore = 0; lcore < vm->avail_cores ; lcore++){\r
+\r
+ PrintDebug(vm, VCORE_NONE,"EDF Sched. edf_sched_init. Initializing logical core %d\n", lcore);\r
+\r
+ struct vm_edf_rq * edf_rq_list = (struct vm_edf_rq *) sched_state->priv_data;\r
+ struct vm_edf_rq * edf_rq = &edf_rq_list[lcore];\r
+ \r
+ edf_rq->vCPUs_tree = RB_ROOT;\r
+ edf_rq->cpu_u=0;\r
+ edf_rq->nr_vCPU=0;\r
+ edf_rq->curr_vCPU=NULL;\r
+ edf_rq->rb_leftmost=NULL;\r
+ edf_rq->last_sched_time=0;\r
+ init_edf_config(&edf_rq->edf_config);\r
+\r
+ }\r
+ \r
+ return 0;\r
+ \r
+}\r
+\r
+\r
+/*\r
+ * is_admissible_core: Decides if a core is admited to the red black tree according with \r
+ * the admisibility formula.\r
+ */\r
+\r
+static bool \r
+is_admissible_core(struct vm_core_edf_sched * new_sched_core, struct vm_edf_rq *runqueue){\r
+\r
+ int curr_utilization = runqueue->cpu_u;\r
+ int new_utilization = curr_utilization + (100 * new_sched_core->slice / new_sched_core->period);\r
+ int cpu_percent = (runqueue->edf_config).cpu_percent; \r
+\r
+ if (new_utilization <= cpu_percent)\r
+ return true;\r
+ else\r
+ return false; \r
+\r
+}\r
+\r
+\r
+/*\r
+ * count_cores: Function useful to count the number of cores in a runqueue (Not used for now)\r
+ *\r
+ */\r
+\r
+\r
+/*static int count_cores(struct vm_edf_rq *runqueue){\r
+\r
+ struct rb_node *node = v3_rb_first(&runqueue->vCPUs_tree);\r
+ struct vm_core_edf_sched *curr_core;\r
+ int number_cores = 0; \r
+\r
+ while(node){\r
+ \r
+ curr_core = container_of(node, struct vm_core_edf_sched, node);\r
+ node = v3_rb_next(node);\r
+ number_cores++;\r
+ }\r
+\r
+ return number_cores;\r
+}*/ \r
+\r
+\r
+\r
+/*\r
+ * insert_core_edf: Finds a place in the tree for a newly activated core, adds the node \r
+ * and rebalaces the tree\r
+ */\r
+\r
+static bool \r
+insert_core_edf(struct vm_core_edf_sched *core, struct vm_edf_rq *runqueue){\r
+\r
+ struct rb_node **new_core = &(runqueue->vCPUs_tree.rb_node);\r
+ struct rb_node *parent = NULL;\r
+ struct vm_core_edf_sched *curr_core;\r
+\r
+ // Find out place in the tree for the new core \r
+ while (*new_core) {\r
+ \r
+ curr_core = container_of(*new_core, struct vm_core_edf_sched, node);\r
+ parent = *new_core;\r
+ \r
+ if (core->current_deadline < curr_core->current_deadline)\r
+ new_core = &((*new_core)->rb_left);\r
+ else if (core->current_deadline > curr_core->current_deadline)\r
+ new_core = &((*new_core)->rb_right);\r
+ else // Is Possible to have same current deadlines in both cores!\r
+ return false;\r
+ }\r
+ // Add new node and rebalance tree. \r
+ rb_link_node(&core->node, parent, new_core);\r
+ v3_rb_insert_color(&core->node, &runqueue->vCPUs_tree);\r
+ \r
+ return true;\r
+ } \r
+\r
+\r
+/*\r
+ * get_curr_host_time: Calculates the current host time (microseconds)\r
+ */\r
+\r
+static uint64_t \r
+get_curr_host_time(struct vm_core_time *core_time){\r
+\r
+ uint64_t cur_cycle = v3_get_host_time(core_time);\r
+ uint64_t cpu_khz = core_time->host_cpu_freq;\r
+ uint64_t curr_time_us = 1000 * cur_cycle / cpu_khz;\r
+\r
+ return curr_time_us;\r
+\r
+}\r
+\r
+\r
+/*\r
+ * next_start_period: Given the current host time and the period of a given vCPU, \r
+ * calculates the time in which its next period starts.\r
+ *\r
+ */\r
+\r
+static uint64_t \r
+next_start_period(uint64_t curr_time_us, uint64_t period_us){\r
+\r
+ uint64_t time_period_us = curr_time_us % period_us;\r
+ uint64_t remaining_time_us = period_us - time_period_us;\r
+ uint64_t next_start_us = curr_time_us + remaining_time_us;\r
+\r
+ return next_start_us;\r
+\r
+}\r
+\r
+/*\r
+ * get_runqueue: Get the runqueue assigned to a virtual core.\r
+ */\r
+\r
+struct vm_edf_rq * get_runqueue(struct guest_info *info){\r
+\r
+ struct vm_edf_rq *runqueue_list = (struct vm_edf_rq *) info->vm_info->sched.priv_data;\r
+ struct vm_edf_rq *runqueue = &runqueue_list[info->pcpu_id]; \r
+ return runqueue;\r
+}\r
+\r
+\r
+/*\r
+ * wakeup_core: Wakeup a given vCPU thread\r
+ */\r
+\r
+static void \r
+wakeup_core(struct guest_info *info){\r
+\r
+ struct vm_core_edf_sched *core = info->core_sched.priv_data;\r
+ struct vm_edf_rq *runqueue = get_runqueue(info);\r
+\r
+ if (!info->core_thread) {\r
+ PrintError(info->vm_info, info,"ERROR: Tried to wakeup non-existent core thread vCPU_id %d \n",info->vcpu_id);\r
+ } \r
+ else {\r
+\r
+ PrintDebug(info->vm_info, info,"EDF Sched. run_next_core. vcpu_id %d, logical id %d, Total time %llu, Miss_deadlines %d, slice_overuses %d extra_time %llu, thread (%p)\n", \r
+ core->info->vcpu_id,\r
+ core->info->pcpu_id,\r
+ core->total_time,\r
+ core->miss_deadline,\r
+ core->slice_overuse,\r
+ core->extra_time_given,\r
+ (struct task_struct *)info->core_thread); \r
+ \r
+ V3_Wakeup(info->core_thread);\r
+ core->last_wakeup_time = get_curr_host_time(&core->info->time_state);\r
+ runqueue->curr_vCPU = core;\r
+\r
+ }\r
+\r
+}\r
+\r
+\r
+/*\r
+ * activate_core - Moves a core to the red-black tree.\r
+ * used time is set to zero and current deadline is calculated \r
+ */\r
+\r
+static void \r
+activate_core(struct vm_core_edf_sched * core, struct vm_edf_rq *runqueue){\r
+ \r
+ if (is_admissible_core(core, runqueue)){\r
+ \r
+ uint64_t curr_time_us = get_curr_host_time(&core->info->time_state);\r
+ uint64_t curr_deadline = next_start_period(curr_time_us, core->period);\r
+ \r
+ core->current_deadline = curr_deadline;\r
+ core->used_time=0; \r
+ core->remaining_time=core->slice; \r
+ \r
+ bool ins = insert_core_edf(core, runqueue);\r
+ /* \r
+ * If not inserted is possible that there is other core with the same deadline.\r
+ * Then, the deadline is modified and try again \r
+ */ \r
+ while(!ins){ \r
+ core->current_deadline ++;\r
+ ins = insert_core_edf(core, runqueue); \r
+ } \r
+ \r
+ runqueue->cpu_u += 100 * core->slice / core->period;\r
+ runqueue->nr_vCPU ++;\r
+ \r
+ /*\r
+ * If this is the first time to be activated pick first earliest deadline core to wakeup.\r
+ */\r
+\r
+ if(core->last_wakeup_time == 0){\r
+\r
+ struct vm_core_edf_sched *next_core;\r
+ \r
+ /*\r
+ * Pick first earliest deadline core\r
+ */\r
+ struct rb_node *node = v3_rb_first(&runqueue->vCPUs_tree);\r
+ next_core = container_of(node, struct vm_core_edf_sched, node);\r
+ \r
+ // Wakeup next_core\r
+ wakeup_core(next_core->info);\r
+ \r
+ //Sleep old core\r
+ \r
+ V3_Sleep(0);\r
+ }\r
+ \r
+ }\r
+ else \r
+ PrintError(core->info->vm_info, core->info,"EDF Sched. activate_core. CPU cannot activate the core. It is not admissible"); \r
+}\r
+\r
+\r
+/*\r
+ * edf_sched_core_init: Initializes per core data structure and \r
+ * calls activate function.\r
+ */\r
+\r
+int \r
+edf_sched_core_init(struct guest_info * info){\r
+\r
+ struct vm_edf_rq *runqueue = get_runqueue(info);\r
+ struct vm_core_edf_sched *core_edf;\r
+\r
+ PrintDebug(info->vm_info, info,"EDF Sched. Initializing vcore %d\n", info->vcpu_id);\r
+\r
+ core_edf = (struct vm_core_edf_sched *) V3_Malloc(sizeof (struct vm_core_edf_sched));\r
+ if (!core_edf) {\r
+ PrintError(info->vm_info, info,"Cannot allocate private_data in edf_sched_core_init\n");\r
+ return -1;\r
+ }\r
+ info->core_sched.priv_data = core_edf;\r
+ \r
+ // Default configuration if not specified in configuration file \r
+ \r
+ core_edf->info = info; \r
+ core_edf->period = 500000;\r
+ core_edf->slice = 50000;\r
+ core_edf->used_time = 0;\r
+ core_edf->last_wakeup_time = 0;\r
+ core_edf->remaining_time = core_edf->slice; \r
+ core_edf->miss_deadline = 0;\r
+ core_edf->extra_time = true;\r
+ core_edf->total_time = 0;\r
+ core_edf->slice_overuse = 0;\r
+ core_edf->extra_time_given = 0;\r
+\r
+ v3_cfg_tree_t * cfg_tree = core_edf->info->vm_info->cfg_data->cfg;\r
+ v3_cfg_tree_t * core = v3_cfg_subtree(v3_cfg_subtree(cfg_tree, "cores"), "core");\r
+ \r
+ while (core){\r
+ char *id = v3_cfg_val(core, "vcpu_id");\r
+ char *period = v3_cfg_val(core, "period");\r
+ char *slice = v3_cfg_val(core, "slice");\r
+ char *extra_time = v3_cfg_val(core, "extra_time");\r
+ \r
+ if (atoi(id) == core_edf->info->vcpu_id){\r
+ \r
+ core_edf->period = atoi(period);\r
+ core_edf->slice = atoi(slice);\r
+ core_edf->remaining_time = core_edf->slice; \r
+ if (strcasecmp(extra_time, "true") == 0)\r
+ core_edf->extra_time = true;\r
+ else \r
+ core_edf->extra_time = false;\r
+ break;\r
+ }\r
+ core = v3_cfg_next_branch(core);\r
+ }\r
+\r
+ activate_core(core_edf,runqueue); \r
+ return 0; \r
+}\r
+\r
+/*\r
+ * search_core_edf: Searches a core in the red-black tree by using its vcpu_id\r
+ */\r
+static struct vm_core_edf_sched * \r
+search_core_edf(struct vm_core_edf_sched *core_edf, struct vm_edf_rq *runqueue){\r
+\r
+ struct rb_node *node = runqueue->vCPUs_tree.rb_node;\r
+ \r
+ while (node) {\r
+ \r
+ struct vm_core_edf_sched *core = container_of(node, struct vm_core_edf_sched, node);\r
+ \r
+ if (core_edf->current_deadline < core->current_deadline)\r
+ node = node->rb_left;\r
+ else if (core_edf->current_deadline > core->current_deadline)\r
+ node = node->rb_right;\r
+ else\r
+ if(core->info->vcpu_id == core_edf->info->vcpu_id){\r
+ return core;\r
+ }\r
+ }\r
+ return NULL;\r
+}\r
+\r
+\r
+/* \r
+ * delete_core_edf: Deletes a core from the red black tree, generally when it has \r
+ * consumed its time slice within the current period.\r
+ */\r
+\r
+static bool \r
+delete_core_edf( struct vm_core_edf_sched *core_edf , struct vm_edf_rq *runqueue){\r
+\r
+ struct vm_core_edf_sched *core = search_core_edf(core_edf, runqueue);\r
+ if (core){ \r
+\r
+ v3_rb_erase(&core->node, &runqueue->vCPUs_tree); \r
+ return true;\r
+ } \r
+ else{\r
+ PrintError(core->info->vm_info, core->info,"EDF Sched. delete_core_edf.Attempted to erase unexisting core");\r
+ return false; \r
+ }\r
+}\r
+\r
+\r
+/*\r
+ * deactivate_core - Removes a core from the red-black tree.\r
+ */\r
+\r
+static void \r
+deactivate_core(struct vm_core_edf_sched * core, struct vm_edf_rq *runqueue){\r
+\r
+ if(delete_core_edf(core, runqueue)){\r
+ runqueue->cpu_u -= 100 * core->slice / core->period;\r
+ runqueue->nr_vCPU -- ;\r
+ } \r
+}\r
+\r
+\r
+/*\r
+ * pick_next_core: Returns the next core to be scheduled from the red black tree\r
+ */\r
+\r
+static struct vm_core_edf_sched * \r
+pick_next_core(struct vm_edf_rq *runqueue){\r
+ \r
+ \r
+ /*\r
+ * Pick first earliest deadline core\r
+ */\r
+ struct rb_node *node = v3_rb_first(&runqueue->vCPUs_tree);\r
+ struct vm_core_edf_sched *next_core = container_of(node, struct vm_core_edf_sched, node);\r
+ \r
+ /* \r
+ * Verify if the earliest deadline core has used its complete slice and return it if not\r
+ */\r
+\r
+ if (next_core->used_time < next_core->slice){\r
+ if(next_core->current_deadline < get_curr_host_time(&next_core->info->time_state))\r
+ next_core->miss_deadline++; \r
+ return next_core;\r
+ }\r
+ /*\r
+ * If slice used, pick the next core that has not used its complete slice \r
+ */\r
+\r
+ else { \r
+ while(next_core->used_time >= next_core->slice){\r
+ \r
+ if(next_core->current_deadline < get_curr_host_time(&next_core->info->time_state) || !next_core->extra_time ){\r
+\r
+ deactivate_core(next_core,runqueue); \r
+ activate_core(next_core,runqueue);\r
+ \r
+ } \r
+\r
+ node = v3_rb_next(node);\r
+ if(node){\r
+ next_core = container_of(node, struct vm_core_edf_sched, node);\r
+ }\r
+ else{ \r
+ node = v3_rb_first(&runqueue->vCPUs_tree); // If all cores have used its slice return the first one\r
+ return container_of(node, struct vm_core_edf_sched, node);\r
+ } \r
+\r
+ }\r
+ }\r
+\r
+ return next_core;\r
+}\r
+\r
+\r
+static void \r
+adjust_slice(struct guest_info * info, int used_time, int extra_time)\r
+{\r
+ struct vm_core_edf_sched *core = info->core_sched.priv_data;\r
+ struct vm_edf_rq *runqueue = get_runqueue(info);\r
+\r
+ core->used_time = used_time;\r
+ \r
+ if (extra_time >= 0) {\r
+ core->used_time += extra_time;\r
+ }\r
+\r
+ if( core->used_time >= core->slice){ \r
+ deactivate_core(core,runqueue);\r
+ activate_core(core,runqueue);\r
+ }\r
+}\r
+\r
+\r
+/*\r
+ * run_next_core: Pick next core to be scheduled and wakeup it\r
+ */\r
+\r
+static void \r
+run_next_core(struct guest_info *info, int used_time, int usec)\r
+{\r
+ struct vm_core_edf_sched *core = info->core_sched.priv_data;\r
+ struct vm_core_edf_sched *next_core;\r
+ struct vm_edf_rq *runqueue = get_runqueue(info);\r
+ \r
+ /* The next core to be scheduled is choosen from the tree (Function pick_next_core). \r
+ * The selected core is the one with the earliest deadline and with available time \r
+ * to use within the current period (used_time < slice) \r
+ */\r
+ \r
+ next_core = pick_next_core(runqueue); // Pick next core to schedule\r
+ \r
+ if (core != next_core){\r
+\r
+ // Wakeup next_core\r
+ wakeup_core(next_core->info);\r
+ core->total_time += used_time;\r
+\r
+ if (used_time > core->slice){\r
+ core->slice_overuse++;\r
+ core->extra_time_given += (used_time - core->slice);\r
+ }\r
+\r
+ // Sleep old core\r
+ \r
+ V3_Sleep(usec);\r
+ \r
+ }\r
+}\r
+\r
+\r
+/*\r
+ * edf_schedule: Scheduling function\r
+ */\r
+\r
+static void\r
+edf_schedule(struct guest_info * info, int usec){\r
+\r
+ uint64_t host_time = get_curr_host_time(&info->time_state);\r
+ struct vm_edf_rq *runqueue = get_runqueue(info); \r
+ struct vm_core_edf_sched *core = (struct vm_core_edf_sched *) info->core_sched.priv_data;\r
+\r
+ uint64_t used_time = 0;\r
+ if(core->last_wakeup_time != 0) \r
+ used_time = host_time - core->last_wakeup_time;\r
+\r
+ if(usec == 0) runqueue->last_sched_time = host_time; // Called from edf_sched_scheduled\r
+ adjust_slice(core->info, host_time - core->last_wakeup_time, usec);\r
+\r
+ run_next_core(core->info,used_time, usec);\r
+ return;\r
+\r
+}\r
+\r
+/*\r
+ * edf_sched_schedule: Main scheduling function. Computes amount of time in period left,\r
+ * recomputing the current core's deadline if it has expired, then runs\r
+ * scheduler \r
+ * It is called in the following cases:\r
+ * A vCPU becomes runnable\r
+ * The slice of the current vCPU was used\r
+ * The period of a vCPU in the runqueue starts\r
+ * Other case?? \r
+ * TODO Something to do with extra time?\r
+ * TODO Check the use of remaining_time\r
+ */\r
+\r
+void \r
+edf_sched_schedule(struct guest_info * info){\r
+\r
+ edf_schedule(info, 0);\r
+ return;\r
+}\r
+\r
+/*\r
+ * edf_sched_yield: Called when yielding the logical cpu for usec is needed\r
+ */\r
+\r
+void \r
+edf_sched_yield(struct guest_info * info, int usec){\r
+ \r
+ edf_schedule(info, usec);\r
+ return;\r
+ \r
+}\r
+\r
+/*\r
+ * edf_sched_deinit: Frees edf scheduler data structures\r
+ */\r
+\r
+\r
+int \r
+edf_sched_deinit(struct v3_vm_info *vm)\r
+{\r
+\r
+ struct vm_scheduler * sched = vm->sched.sched;\r
+ void *priv_data = vm->sched.priv_data;\r
+ \r
+ if (sched) \r
+ V3_Free(sched); \r
+\r
+ if (priv_data) \r
+ V3_Free(priv_data);\r
+\r
+ return 0;\r
+\r
+}\r
+\r
+/*\r
+ * edf_sched_deinit: Frees virtual core data structures\r
+ */\r
+\r
+int \r
+edf_sched_core_deinit(struct guest_info *core)\r
+{\r
+\r
+ struct vm_scheduler * sched = core->core_sched.sched;\r
+ void *priv_data = core->core_sched.priv_data;\r
+ \r
+ if (sched) \r
+ V3_Free(sched); \r
+\r
+ if (priv_data) \r
+ V3_Free(priv_data);\r
+\r
+ return 0;\r
+}\r
+\r
+static struct vm_scheduler_impl edf_sched = {\r
+ .name = "edf",\r
+ .init = edf_sched_init,\r
+ .deinit = edf_sched_deinit,\r
+ .core_init = edf_sched_core_init,\r
+ .core_deinit = edf_sched_core_deinit,\r
+ .schedule = edf_sched_schedule,\r
+ .yield = edf_sched_yield\r
+};\r
+\r
+static int \r
+ext_sched_edf_init() {\r
+ \r
+ PrintDebug(VM_NONE, VCORE_NONE,"Sched. Creating (%s) scheduler\n",edf_sched.name);\r
+ return v3_register_scheduler(&edf_sched);\r
+}\r
+\r
+static int \r
+ext_sched_edf_vm_init() {\r
+ return 0;\r
+}\r
+\r
+static struct v3_extension_impl sched_edf_impl = {\r
+ .name = "EDF Scheduler",\r
+ .init = ext_sched_edf_init,\r
+ .vm_init = ext_sched_edf_vm_init,\r
+ .vm_deinit = NULL,\r
+ .core_init = NULL,\r
+ .core_deinit = NULL,\r
+ .on_entry = NULL,\r
+ .on_exit = NULL\r
+};\r
+\r
+register_extension(&sched_edf_impl);\r
--- /dev/null
+/*
+ * This file is part of the Palacios Virtual Machine Monitor developed
+ * by the V3VEE Project with funding from the United States National
+ * Science Foundation and the Department of Energy.
+ *
+ * The V3VEE Project is a joint project between Northwestern University
+ * and the University of New Mexico. You can find out more at
+ * http://www.v3vee.org
+ *
+ * Copyright (c) 2008, Jack Lange <jarusl@cs.northwestern.edu>
+ * Copyright (c) 2008, The V3VEE Project <http://www.v3vee.org>
+ * All rights reserved.
+ *
+ * Author: Oscar Mondragon <omondrag@cs.unm.edu>
+ * Patrick G. Bridges <bridges@cs.unm.edu>
+ *
+ * This is free software. You are permitted to use,
+ * redistribute, and modify it as specified in the file "V3VEE_LICENSE".
+ */
+
+#include <palacios/vmm.h>
+#include <palacios/vm_guest.h>
+#include <palacios/vmm_scheduler.h>
+#include <palacios/vmm_hashtable.h>
+
+#ifndef V3_CONFIG_DEBUG_SCHEDULER
+#undef PrintDebug
+#define PrintDebug(fmt, args...)
+#endif
+
+static char default_strategy[] = "host";
+static struct hashtable * master_scheduler_table = NULL;
+static int create_host_scheduler();
+
+static struct vm_scheduler_impl *scheduler = NULL;
+
+static uint_t scheduler_hash_fn(addr_t key) {
+ char * name = (char *)key;
+ return v3_hash_buffer((uint8_t *)name, strlen(name));
+}
+
+static int scheduler_eq_fn(addr_t key1, addr_t key2) {
+ char * name1 = (char *)key1;
+ char * name2 = (char *)key2;
+
+ return (strcmp(name1, name2) == 0);
+}
+
+int V3_init_scheduling() {
+
+ PrintDebug(VM_NONE, VCORE_NONE,"Initializing scheduler");
+
+ master_scheduler_table = v3_create_htable(0, scheduler_hash_fn, scheduler_eq_fn);
+ return create_host_scheduler();
+}
+
+
+int v3_register_scheduler(struct vm_scheduler_impl *s) {
+
+ PrintDebug(VM_NONE, VCORE_NONE,"Registering Scheduler (%s)\n", s->name);
+
+ if (v3_htable_search(master_scheduler_table, (addr_t)(s->name))) {
+ PrintError(VM_NONE, VCORE_NONE, "Multiple instances of scheduler (%s)\n", s->name);
+ return -1;
+ }
+ PrintDebug(VM_NONE, VCORE_NONE,"Registering Scheduler (%s) 2\n", s->name);
+
+
+ if (v3_htable_insert(master_scheduler_table,
+ (addr_t)(s->name),
+ (addr_t)(s)) == 0) {
+ PrintError(VM_NONE, VCORE_NONE, "Could not register scheduler (%s)\n", s->name);
+ return -1;
+ }
+
+ PrintDebug(VM_NONE, VCORE_NONE,"Scheduler registered\n");
+ return 0;
+}
+
+struct vm_scheduler_impl *v3_scheduler_lookup(char *name)
+{
+ return (struct vm_scheduler_impl *)v3_htable_search(master_scheduler_table, (addr_t)(name));
+}
+
+int V3_enable_scheduler() {
+ /* XXX Lookup the specified scheduler to use for palacios and use it */
+ scheduler = v3_scheduler_lookup(default_strategy);
+ if (!scheduler) {
+ PrintError(VM_NONE, VCORE_NONE,"Specified Palacios scheduler \"%s\" not found.\n", default_strategy);
+ return -1;
+ }
+ if (scheduler->init) {
+ return scheduler->init();
+ } else {
+ return 0;
+ }
+}
+
+int v3_scheduler_register_vm(struct v3_vm_info *vm) {
+ if (scheduler->vm_init) {
+ return scheduler->vm_init(vm);
+ } else {
+ return 0;
+ }
+}
+int v3_scheduler_register_core(struct guest_info *core) {
+ if (scheduler->core_init) {
+ return scheduler->core_init(core);
+ } else {
+ return 0;
+ }
+}
+int v3_scheduler_admit_vm(struct v3_vm_info *vm) {
+ if (scheduler->admit) {
+ return scheduler->admit(vm);
+ } else {
+ return 0;
+ }
+}
+int v3_scheduler_notify_remap(struct v3_vm_info *vm) {
+ if (scheduler->remap) {
+ return scheduler->remap(vm);
+ } else {
+ return 0;
+ }
+}
+int v3_scheduler_notify_dvfs(struct v3_vm_info *vm) {
+ if (scheduler->dvfs) {
+ return scheduler->dvfs(vm);
+ } else {
+ return 0;
+ }
+}
+void v3_schedule(struct guest_info *core) {
+ if (scheduler->schedule) {
+ scheduler->schedule(core);
+ }
+ return;
+}
+void v3_yield(struct guest_info *core, int usec) {
+ if (scheduler->yield) {
+ scheduler->yield(core, usec);
+ }
+ return;
+}
+
+int host_sched_vm_init(struct v3_vm_info *vm)
+{
+
+ PrintDebug(vm, VCORE_NONE,"Sched. host_sched_init\n");
+
+ char * schedule_hz_str = v3_cfg_val(vm->cfg_data->cfg, "schedule_hz");
+ uint32_t sched_hz = 100;
+
+
+ if (schedule_hz_str) {
+ sched_hz = atoi(schedule_hz_str);
+ }
+
+ PrintDebug(vm, VCORE_NONE,"CPU_KHZ = %d, schedule_freq=%p\n", V3_CPU_KHZ(),
+ (void *)(addr_t)sched_hz);
+
+ uint64_t yield_cycle_period = (V3_CPU_KHZ() * 1000) / sched_hz;
+ vm->sched_priv_data = (void *)yield_cycle_period;
+
+ return 0;
+}
+
+int host_sched_core_init(struct guest_info *core)
+{
+ PrintDebug(core->vm_info, core,"Sched. host_sched_core_init\n");
+
+ uint64_t t = v3_get_host_time(&core->time_state);
+ core->sched_priv_data = (void *)t;
+
+ return 0;
+}
+
+void host_sched_schedule(struct guest_info *core)
+{
+ uint64_t cur_cycle;
+ cur_cycle = v3_get_host_time(&core->time_state);
+
+ if (cur_cycle > ( (uint64_t)core->sched_priv_data + (uint64_t)core->vm_info->sched_priv_data)) {
+
+ V3_Yield();
+
+ uint64_t yield_start_cycle = (uint64_t) core->sched_priv_data;
+ yield_start_cycle += (uint64_t)core->vm_info->sched_priv_data;
+ core->sched_priv_data = (void *)yield_start_cycle;
+
+ }
+}
+
+/*
+ * unconditional cpu yield
+ * if the yielding thread is a guest context, the guest quantum is reset on resumption
+ * Non guest context threads should call this function with a NULL argument
+ *
+ * usec <0 => the non-timed yield is used
+ * usec >=0 => the timed yield is used, which also usually implies interruptible
+ */
+void host_sched_yield(struct guest_info * core, int usec) {
+ uint64_t yield_start_cycle;
+ if (usec < 0) {
+ V3_Yield();
+ } else {
+ V3_Sleep(usec);
+ }
+ yield_start_cycle = (uint64_t) core->sched_priv_data
+ + (uint64_t)core->vm_info->sched_priv_data;
+ core->sched_priv_data = (void *)yield_start_cycle;
+}
+
+
+int host_sched_admit(struct v3_vm_info *vm){
+ return 0;
+}
+
+static struct vm_scheduler_impl host_sched_impl = {
+ .name = "host",
+ .init = NULL,
+ .deinit = NULL,
+ .vm_init = host_sched_vm_init,
+ .vm_deinit = NULL,
+ .core_init = host_sched_core_init,
+ .core_deinit = NULL,
+ .schedule = host_sched_schedule,
+ .yield = host_sched_yield,
+ .admit = host_sched_admit,
+ .remap = NULL,
+ .dvfs=NULL
+};
+
+static int create_host_scheduler()
+{
+ v3_register_scheduler(&host_sched_impl);
+ return 0;
+}
