* 1. Queue locks. Actual irq insertions are done via queueing irq ops at the dest apic.
* The destination apic's core is responsible for draining the queue, and actually
* setting the vector table.
- * 2. State locks. This is a standard lock taken when internal apic state is read/written.
+ * 2. State lock. This is a standard lock taken when internal apic state is read/written.
* When an irq's destination is determined this lock is taken to examine the apic's
* addressability.
* 3. VM barrier lock. This is taken when actual VM core state is changed (via SIPI).
+
typedef enum {INIT_ST,
SIPI,
STARTED} ipi_state_t;
struct v3_timer * timer;
- v3_lock_t state_lock;
+
struct v3_queue irq_queue;
uint32_t eoi;
struct apic_dev_state {
int num_apics;
+ v3_lock_t state_lock;
+
struct apic_state apics[0];
} __attribute__((packed));
}
+
+
+
// irq_num is the bit offset into a 256 bit buffer...
static int activate_apic_irq(struct apic_state * apic, uint32_t irq_num) {
int major_offset = (irq_num & ~0x00000007) >> 3;
uint8_t flag = 0x1 << minor_offset;
- if (irq_num <= 15 || irq_num > 255) {
- PrintError("apic %u: core %d: Attempting to raise an invalid interrupt: %d\n",
- apic->lapic_id.val, apic->core->vcpu_id, irq_num);
- return -1;
- }
-
-
PrintDebug("apic %u: core %d: Raising APIC IRQ %d\n", apic->lapic_id.val, apic->core->vcpu_id, irq_num);
if (*req_location & flag) {
}
+static int add_apic_irq_entry(struct apic_state * apic, uint8_t irq_num) {
+
+ if (irq_num <= 15) {
+ PrintError("core %d: Attempting to raise an invalid interrupt: %d\n",
+ apic->core->vcpu_id, irq_num);
+ return -1;
+ }
+
+ v3_enqueue(&(apic->irq_queue), (addr_t)irq_num);
+
+ return 0;
+}
+
+static void drain_irq_entries(struct apic_state * apic) {
+ uint32_t irq = 0;
+
+ while ((irq = (uint32_t)v3_dequeue(&(apic->irq_queue))) != 0) {
+ activate_apic_irq(apic, irq);
+ }
+
+}
+
+
+
static int get_highest_isr(struct apic_state * apic) {
int i = 0, j = 0;
if (del_mode == APIC_FIXED_DELIVERY) {
//PrintDebug("Activating internal APIC IRQ %d\n", vec_num);
- return activate_apic_irq(apic, vec_num);
+ return add_apic_irq_entry(apic, vec_num);
} else {
PrintError("apic %u: core ?: Unhandled Delivery Mode\n", apic->lapic_id.val);
return -1;
-static inline int should_deliver_cluster_ipi(struct guest_info * dst_core,
+static inline int should_deliver_cluster_ipi(struct apic_dev_state * apic_dev,
+ struct guest_info * dst_core,
struct apic_state * dst_apic, uint8_t mda) {
+ int ret = 0;
+
+
if ( ((mda & 0xf0) == (dst_apic->log_dst.dst_log_id & 0xf0)) && /* (I am in the cluster and */
((mda & 0x0f) & (dst_apic->log_dst.dst_log_id & 0x0f)) ) { /* I am in the set) */
+ ret = 1;
+ } else {
+ ret = 0;
+ }
+
+ if (ret == 1) {
PrintDebug("apic %u core %u: accepting clustered IRQ (mda 0x%x == log_dst 0x%x)\n",
dst_apic->lapic_id.val, dst_core->vcpu_id, mda,
dst_apic->log_dst.dst_log_id);
-
- return 1;
} else {
PrintDebug("apic %u core %u: rejecting clustered IRQ (mda 0x%x != log_dst 0x%x)\n",
dst_apic->lapic_id.val, dst_core->vcpu_id, mda,
dst_apic->log_dst.dst_log_id);
- return 0;
}
+
+ return ret;
+
}
-static inline int should_deliver_flat_ipi(struct guest_info * dst_core,
+static inline int should_deliver_flat_ipi(struct apic_dev_state * apic_dev,
+ struct guest_info * dst_core,
struct apic_state * dst_apic, uint8_t mda) {
- if (dst_apic->log_dst.dst_log_id & mda) { // I am in the set
+ int ret = 0;
- PrintDebug("apic %u core %u: accepting flat IRQ (mda 0x%x == log_dst 0x%x)\n",
- dst_apic->lapic_id.val, dst_core->vcpu_id, mda,
- dst_apic->log_dst.dst_log_id);
- return 1;
+ if ((dst_apic->log_dst.dst_log_id & mda) != 0) { // I am in the set
+ ret = 1;
+ } else {
+ ret = 0;
+ }
- } else {
+ if (ret == 1) {
+ PrintDebug("apic %u core %u: accepting flat IRQ (mda 0x%x == log_dst 0x%x)\n",
+ dst_apic->lapic_id.val, dst_core->vcpu_id, mda,
+ dst_apic->log_dst.dst_log_id);
+ } else {
PrintDebug("apic %u core %u: rejecting flat IRQ (mda 0x%x != log_dst 0x%x)\n",
dst_apic->lapic_id.val, dst_core->vcpu_id, mda,
dst_apic->log_dst.dst_log_id);
- return 0;
- }
+ }
+
+
+ return ret;
}
-static int should_deliver_ipi(struct guest_info * dst_core,
+static int should_deliver_ipi(struct apic_dev_state * apic_dev,
+ struct guest_info * dst_core,
struct apic_state * dst_apic, uint8_t mda) {
+ addr_t flags = 0;
+ int ret = 0;
+ flags = v3_lock_irqsave(apic_dev->state_lock);
if (dst_apic->dst_fmt.model == 0xf) {
if (mda == 0xff) {
/* always deliver broadcast */
- return 1;
+ ret = 1;
+ } else {
+ ret = should_deliver_flat_ipi(apic_dev, dst_core, dst_apic, mda);
}
-
- return should_deliver_flat_ipi(dst_core, dst_apic, mda);
-
} else if (dst_apic->dst_fmt.model == 0x0) {
if (mda == 0xff) {
/* always deliver broadcast */
- return 1;
+ ret = 1;
+ } else {
+ ret = should_deliver_cluster_ipi(apic_dev, dst_core, dst_apic, mda);
}
- return should_deliver_cluster_ipi(dst_core, dst_apic, mda);
-
} else {
+ ret = -1;
+ }
+
+ v3_unlock_irqrestore(apic_dev->state_lock, flags);
+
+
+ if (ret == -1) {
PrintError("apic %u core %u: invalid destination format register value 0x%x for logical mode delivery.\n",
dst_apic->lapic_id.val, dst_core->vcpu_id, dst_apic->dst_fmt.model);
- return -1;
}
+
+ return ret;
}
+
+
// Only the src_apic pointer is used
static int deliver_ipi(struct apic_state * src_apic,
struct apic_state * dst_apic,
// lowest priority -
// caller needs to have decided which apic to deliver to!
- int do_xcall;
-
PrintDebug("delivering IRQ %d to core %u\n", vector, dst_core->vcpu_id);
- do_xcall = activate_apic_irq(dst_apic, vector);
+ add_apic_irq_entry(dst_apic, vector);
-
-
+#ifdef V3_CONFIG_MULTITHREAD_OS
if (dst_apic != src_apic) {
PrintDebug(" non-local core with new interrupt, forcing it to exit now\n");
-
-#ifdef V3_CONFIG_MULTITHREAD_OS
v3_interrupt_cpu(dst_core->vm_info, dst_core->pcpu_id, 0);
-#endif
}
+#endif
+
break;
}
break;
}
- // Write the RIP, CS, and descriptor
- // assume the rest is already good to go
- //
- // vector VV -> rip at 0
- // CS = VV00
- // This means we start executing at linear address VV000
- //
- // So the selector needs to be VV00
- // and the base needs to be VV000
- //
- dst_core->rip = 0;
- dst_core->segments.cs.selector = vector << 8;
- dst_core->segments.cs.limit = 0xffff;
- dst_core->segments.cs.base = vector << 12;
+ v3_reset_vm_core(dst_core, vector);
PrintDebug(" SIPI delivery (0x%x -> 0x%x:0x0) to core %u\n",
vector, dst_core->segments.cs.selector, dst_core->vcpu_id);
break;
}
+
+ case APIC_EXTINT_DELIVERY: // EXTINT
+ /* Two possible things to do here:
+ * 1. Ignore the IPI and assume the 8259a (PIC) will handle it
+ * 2. Add 32 to the vector and inject it...
+ * We probably just want to do 1 here, and assume the raise_irq() will hit the 8259a.
+ */
+ return 0;
+
case APIC_SMI_DELIVERY:
case APIC_RES1_DELIVERY: // reserved
case APIC_NMI_DELIVERY:
- case APIC_EXTINT_DELIVERY: // ExtInt
default:
PrintError("IPI %d delivery is unsupported\n", del_mode);
return -1;
}
-static struct apic_state * find_physical_apic(struct apic_dev_state *apic_dev, struct int_cmd_reg *icr)
-{
+static struct apic_state * find_physical_apic(struct apic_dev_state * apic_dev, uint32_t dst_idx) {
+ struct apic_state * dst_apic = NULL;
+ addr_t flags;
int i;
-
- if ( (icr->dst > 0) && (icr->dst < apic_dev->num_apics) ) {
+
+ flags = v3_lock_irqsave(apic_dev->state_lock);
+
+ if ( (dst_idx > 0) && (dst_idx < apic_dev->num_apics) ) {
// see if it simply is the core id
- if (apic_dev->apics[icr->dst].lapic_id.val == icr->dst) {
- return &(apic_dev->apics[icr->dst]);
+ if (apic_dev->apics[dst_idx].lapic_id.val == dst_idx) {
+ dst_apic = &(apic_dev->apics[dst_idx]);
}
}
for (i = 0; i < apic_dev->num_apics; i++) {
- if (apic_dev->apics[i].lapic_id.val == icr->dst) {
- return &(apic_dev->apics[i]);
+ if (apic_dev->apics[i].lapic_id.val == dst_idx) {
+ dst_apic = &(apic_dev->apics[i]);
}
}
-
- return NULL;
+
+ v3_unlock_irqrestore(apic_dev->state_lock, flags);
+
+ return dst_apic;
}
case APIC_SHORTHAND_NONE: // no shorthand
if (icr->dst_mode == APIC_DEST_PHYSICAL) {
- dest_apic = find_physical_apic(apic_dev, icr);
+ dest_apic = find_physical_apic(apic_dev, icr->dst);
if (dest_apic == NULL) {
PrintError("apic: Attempted send to unregistered apic id=%u\n", icr->dst);
dest_apic = &(apic_dev->apics[i]);
- del_flag = should_deliver_ipi(dest_apic->core, dest_apic, mda);
+ del_flag = should_deliver_ipi(apic_dev, dest_apic->core, dest_apic, mda);
if (del_flag == -1) {
}
}
}
- } else { //APIC_LOWEST_DELIVERY
- int i;
+ } else { // APIC_LOWEST_DELIVERY
struct apic_state * cur_best_apic = NULL;
uint8_t mda = icr->dst;
-
+ int i;
+
// logical, lowest priority
for (i = 0; i < apic_dev->num_apics; i++) {
dest_apic = &(apic_dev->apics[i]);
- del_flag = should_deliver_ipi(dest_apic->core, dest_apic, mda);
+ del_flag = should_deliver_ipi(apic_dev, dest_apic->core, dest_apic, mda);
if (del_flag == -1) {
PrintError("apic: Error checking delivery mode\n");
return -1;
} else if (del_flag == 1) {
// update priority for lowest priority scan
+ addr_t flags = 0;
+
+ flags = v3_lock_irqsave(apic_dev->state_lock);
+
if (cur_best_apic == 0) {
cur_best_apic = dest_apic;
} else if (dest_apic->task_prio.val < cur_best_apic->task_prio.val) {
cur_best_apic = dest_apic;
}
+
+ v3_unlock_irqrestore(apic_dev->state_lock, flags);
+
}
}
*val_ptr = *(((uint8_t *)&val) + byte_addr);
} else if ((length == 2) &&
- ((reg_addr & 0x3) == 0x3)) {
+ ((reg_addr & 0x3) != 0x3)) {
uint_t byte_addr = reg_addr & 0x3;
uint16_t * val_ptr = (uint16_t *)dst;
*val_ptr = *(((uint16_t *)&val) + byte_addr);
addr_t reg_addr = guest_addr - apic->base_addr;
struct apic_msr * msr = (struct apic_msr *)&(apic->base_addr_msr.value);
uint32_t op_val = *(uint32_t *)src;
+ addr_t flags = 0;
PrintDebug("apic %u: core %u: at %p and priv_data is at %p\n",
apic->lapic_id.val, core->vcpu_id, apic, priv_data);
case LDR_OFFSET:
PrintDebug("apic %u: core %u: setting log_dst.val to 0x%x\n",
apic->lapic_id.val, core->vcpu_id, op_val);
+ flags = v3_lock_irqsave(apic_dev->state_lock);
apic->log_dst.val = op_val;
+ v3_unlock_irqrestore(apic_dev->state_lock, flags);
break;
case DFR_OFFSET:
+ flags = v3_lock_irqsave(apic_dev->state_lock);
apic->dst_fmt.val = op_val;
+ v3_unlock_irqrestore(apic_dev->state_lock, flags);
break;
case SPURIOUS_INT_VEC_OFFSET:
apic->spurious_int.val = op_val;
apic->tmr_cur_cnt = op_val;
break;
case TMR_DIV_CFG_OFFSET:
+ PrintDebug("apic %u: core %u: setting tmr_div_cfg to 0x%x\n",
+ apic->lapic_id.val, core->vcpu_id, op_val);
apic->tmr_div_cfg.val = op_val;
break;
static int apic_intr_pending(struct guest_info * core, void * private_data) {
struct apic_dev_state * apic_dev = (struct apic_dev_state *)(private_data);
struct apic_state * apic = &(apic_dev->apics[core->vcpu_id]);
+ int req_irq = 0;
+ int svc_irq = 0;
- // drain irq QUEUE
+ // Activate all queued IRQ entries
+ drain_irq_entries(apic);
- int req_irq = get_highest_irr(apic);
- int svc_irq = get_highest_isr(apic);
+ // Check for newly activated entries
+ req_irq = get_highest_irr(apic);
+ svc_irq = get_highest_isr(apic);
// PrintDebug("apic %u: core %u: req_irq=%d, svc_irq=%d\n",apic->lapic_id.val,info->vcpu_id,req_irq,svc_irq);
struct apic_dev_state * apic_dev = (struct apic_dev_state *)
(((struct vm_device*)dev_data)->private_data);
struct apic_state * apic = &(apic_dev->apics[dst]);
- int do_xcall;
PrintDebug("apic %u core ?: raising interrupt IRQ %u (dst = %u).\n", apic->lapic_id.val, irq, dst);
- do_xcall = activate_apic_irq(apic, irq);
+ add_apic_irq_entry(apic, irq);
- if (do_xcall < 0) {
- PrintError("Failed to activate apic irq\n");
- return -1;
- }
-
- if (do_xcall > 0 && (V3_Get_CPU() != dst)) {
-#ifdef V3_CONFIG_MULTITHREAD_OS
+#ifdef V3_CONFIG_MULTITHREAD_OS
+ if ((V3_Get_CPU() != dst)) {
v3_interrupt_cpu(vm, dst, 0);
-#else
- V3_ASSERT(0);
-#endif
-
}
+#endif
return 0;
}
}
if (apic->tmr_vec_tbl.tmr_mode == APIC_TMR_PERIODIC) {
+ static unsigned int nexits = 0;
+ static unsigned int missed_ints = 0;
+
+ nexits++;
+ missed_ints += tmr_ticks / apic->tmr_init_cnt;
+
+ if ((missed_ints > 0) && (nexits >= 5000)) {
+ V3_Print("apic %u: core %u: missed %u timer interrupts total in last %u exits.\n",
+ apic->lapic_id.val, core->vcpu_id, missed_ints, nexits);
+ missed_ints = 0;
+ nexits = 0;
+ }
+
tmr_ticks = tmr_ticks % apic->tmr_init_cnt;
apic->tmr_cur_cnt = apic->tmr_init_cnt - tmr_ticks;
}
return 0;
}
+#ifdef V3_CONFIG_CHECKPOINT
+static int apic_save(struct v3_chkpt_ctx * ctx, void * private_data) {
+ struct apic_dev_state * apic_state = (struct apic_dev_state *)private_data;
+ int i = 0;
+
+ V3_CHKPT_STD_SAVE(ctx, apic_state->num_apics);
+
+ //V3_CHKPT_STD_SAVE(ctx,apic_state->state_lock);
+ for (i = 0; i < apic_state->num_apics; i++) {
+
+ V3_CHKPT_STD_SAVE(ctx, apic_state->apics[i].base_addr);
+ V3_CHKPT_STD_SAVE(ctx, apic_state->apics[i].base_addr_msr);
+ V3_CHKPT_STD_SAVE(ctx, apic_state->apics[i].lapic_id);
+ V3_CHKPT_STD_SAVE(ctx, apic_state->apics[i].apic_ver);
+ V3_CHKPT_STD_SAVE(ctx, apic_state->apics[i].ext_apic_ctrl);
+ V3_CHKPT_STD_SAVE(ctx, apic_state->apics[i].local_vec_tbl);
+ V3_CHKPT_STD_SAVE(ctx, apic_state->apics[i].tmr_vec_tbl);
+ V3_CHKPT_STD_SAVE(ctx, apic_state->apics[i].tmr_div_cfg);
+ V3_CHKPT_STD_SAVE(ctx, apic_state->apics[i].lint0_vec_tbl);
+ V3_CHKPT_STD_SAVE(ctx, apic_state->apics[i].lint1_vec_tbl);
+ V3_CHKPT_STD_SAVE(ctx, apic_state->apics[i].perf_ctr_loc_vec_tbl);
+ V3_CHKPT_STD_SAVE(ctx, apic_state->apics[i].therm_loc_vec_tbl);
+ V3_CHKPT_STD_SAVE(ctx, apic_state->apics[i].err_vec_tbl);
+ V3_CHKPT_STD_SAVE(ctx, apic_state->apics[i].err_status);
+ V3_CHKPT_STD_SAVE(ctx, apic_state->apics[i].spurious_int);
+ V3_CHKPT_STD_SAVE(ctx, apic_state->apics[i].int_cmd);
+ V3_CHKPT_STD_SAVE(ctx, apic_state->apics[i].log_dst);
+ V3_CHKPT_STD_SAVE(ctx, apic_state->apics[i].dst_fmt);
+ V3_CHKPT_STD_SAVE(ctx, apic_state->apics[i].arb_prio);
+ V3_CHKPT_STD_SAVE(ctx, apic_state->apics[i].task_prio);
+ V3_CHKPT_STD_SAVE(ctx, apic_state->apics[i].proc_prio);
+ V3_CHKPT_STD_SAVE(ctx, apic_state->apics[i].ext_apic_feature);
+ V3_CHKPT_STD_SAVE(ctx, apic_state->apics[i].spec_eoi);
+ V3_CHKPT_STD_SAVE(ctx, apic_state->apics[i].tmr_cur_cnt);
+ V3_CHKPT_STD_SAVE(ctx, apic_state->apics[i].tmr_init_cnt);
+ V3_CHKPT_STD_SAVE(ctx, apic_state->apics[i].ext_intr_vec_tbl);
+ V3_CHKPT_STD_SAVE(ctx, apic_state->apics[i].rem_rd_data);
+ V3_CHKPT_STD_SAVE(ctx, apic_state->apics[i].ipi_state);
+ V3_CHKPT_STD_SAVE(ctx, apic_state->apics[i].int_req_reg);
+ V3_CHKPT_STD_SAVE(ctx, apic_state->apics[i].int_svc_reg);
+ V3_CHKPT_STD_SAVE(ctx, apic_state->apics[i].int_en_reg);
+ V3_CHKPT_STD_SAVE(ctx, apic_state->apics[i].trig_mode_reg);
+ V3_CHKPT_STD_SAVE(ctx, apic_state->apics[i].eoi);
+
+ }
+
+ return 0;
+}
+
+static int apic_load(struct v3_chkpt_ctx * ctx, void * private_data) {
+ struct apic_dev_state *apic_state = (struct apic_dev_state *)private_data;
+ int i = 0;
+
+ V3_CHKPT_STD_LOAD(ctx,apic_state->num_apics);
+
+ for (i = 0; i < apic_state->num_apics; i++) {
+ V3_CHKPT_STD_LOAD(ctx, apic_state->apics[i].base_addr);
+ V3_CHKPT_STD_LOAD(ctx, apic_state->apics[i].base_addr_msr);
+ V3_CHKPT_STD_LOAD(ctx, apic_state->apics[i].lapic_id);
+ V3_CHKPT_STD_LOAD(ctx, apic_state->apics[i].apic_ver);
+ V3_CHKPT_STD_LOAD(ctx, apic_state->apics[i].ext_apic_ctrl);
+ V3_CHKPT_STD_LOAD(ctx, apic_state->apics[i].local_vec_tbl);
+ V3_CHKPT_STD_LOAD(ctx, apic_state->apics[i].tmr_vec_tbl);
+ V3_CHKPT_STD_LOAD(ctx, apic_state->apics[i].tmr_div_cfg);
+ V3_CHKPT_STD_LOAD(ctx, apic_state->apics[i].lint0_vec_tbl);
+ V3_CHKPT_STD_LOAD(ctx, apic_state->apics[i].lint1_vec_tbl);
+ V3_CHKPT_STD_LOAD(ctx, apic_state->apics[i].perf_ctr_loc_vec_tbl);
+ V3_CHKPT_STD_LOAD(ctx, apic_state->apics[i].therm_loc_vec_tbl);
+ V3_CHKPT_STD_LOAD(ctx, apic_state->apics[i].err_vec_tbl);
+ V3_CHKPT_STD_LOAD(ctx, apic_state->apics[i].err_status);
+ V3_CHKPT_STD_LOAD(ctx, apic_state->apics[i].spurious_int);
+ V3_CHKPT_STD_LOAD(ctx, apic_state->apics[i].int_cmd);
+ V3_CHKPT_STD_LOAD(ctx, apic_state->apics[i].log_dst);
+ V3_CHKPT_STD_LOAD(ctx, apic_state->apics[i].dst_fmt);
+ V3_CHKPT_STD_LOAD(ctx, apic_state->apics[i].arb_prio);
+ V3_CHKPT_STD_LOAD(ctx, apic_state->apics[i].task_prio);
+ V3_CHKPT_STD_LOAD(ctx, apic_state->apics[i].proc_prio);
+ V3_CHKPT_STD_LOAD(ctx, apic_state->apics[i].ext_apic_feature);
+ V3_CHKPT_STD_LOAD(ctx, apic_state->apics[i].spec_eoi);
+ V3_CHKPT_STD_LOAD(ctx, apic_state->apics[i].tmr_cur_cnt);
+ V3_CHKPT_STD_LOAD(ctx, apic_state->apics[i].tmr_init_cnt);
+ V3_CHKPT_STD_LOAD(ctx, apic_state->apics[i].ext_intr_vec_tbl);
+ V3_CHKPT_STD_LOAD(ctx, apic_state->apics[i].rem_rd_data);
+ V3_CHKPT_STD_LOAD(ctx, apic_state->apics[i].ipi_state);
+ V3_CHKPT_STD_LOAD(ctx, apic_state->apics[i].int_req_reg);
+ V3_CHKPT_STD_LOAD(ctx, apic_state->apics[i].int_svc_reg);
+ V3_CHKPT_STD_LOAD(ctx, apic_state->apics[i].int_en_reg);
+ V3_CHKPT_STD_LOAD(ctx, apic_state->apics[i].trig_mode_reg);
+ V3_CHKPT_STD_LOAD(ctx, apic_state->apics[i].eoi);
+ }
+
+
+ return 0;
+}
+
+#endif
static struct v3_device_ops dev_ops = {
.free = (int (*)(void *))apic_free,
+#ifdef V3_CONFIG_CHECKPOINT
+ .save = apic_save,
+ .load = apic_load
+#endif
};
sizeof(struct apic_state) * vm->num_cores);
apic_dev->num_apics = vm->num_cores;
+ v3_lock_init(&(apic_dev->state_lock));
struct vm_device * dev = v3_add_device(vm, dev_id, &dev_ops, apic_dev);
