#include <devices/apic.h>
#include <devices/apic_regs.h>
+#include <devices/icc_bus.h>
#include <palacios/vmm.h>
#include <palacios/vmm_msr.h>
+#include <palacios/vmm_sprintf.h>
+#include <palacios/vm_guest.h>
#ifndef CONFIG_DEBUG_APIC
+
+
struct apic_msr {
union {
- uint64_t val;
+ uint64_t value;
struct {
uchar_t rsvd;
uint_t bootstrap_cpu : 1;
} __attribute__((packed));
-
-
struct apic_state {
addr_t base_addr;
/* MSRs */
- v3_msr_t base_addr_msr;
+ struct apic_msr base_addr_msr;
/* memory map registers */
uint32_t eoi;
+ struct vm_device * icc_bus;
+ v3_lock_t lock;
};
-static int apic_read(addr_t guest_addr, void * dst, uint_t length, void * priv_data);
-static int apic_write(addr_t guest_addr, void * src, uint_t length, void * priv_data);
-static void init_apic_state(struct apic_state * apic) {
+
+
+
+static int apic_read(struct guest_info * core, addr_t guest_addr, void * dst, uint_t length, void * priv_data);
+static int apic_write(struct guest_info * core, addr_t guest_addr, void * src, uint_t length, void * priv_data);
+
+static void init_apic_state(struct apic_state * apic, uint32_t id, struct vm_device * icc) {
apic->base_addr = DEFAULT_BASE_ADDR;
- apic->base_addr_msr.value = 0x0000000000000900LL;
+ if (id==0) {
+ // boot processor, enabled
+ apic->base_addr_msr.value = 0x0000000000000900LL;
+ } else {
+ // ap processor, enabled
+ apic->base_addr_msr.value = 0x0000000000000800LL;
+ }
+
+ // same base address regardless of ap or main
apic->base_addr_msr.value |= ((uint64_t)DEFAULT_BASE_ADDR);
- PrintDebug("Sizeof Interrupt Request Register %d, should be 32\n",
- (uint_t)sizeof(apic->int_req_reg));
+ PrintDebug("apic %u: (init_apic_state): msr=0x%llx\n",id, apic->base_addr_msr.value);
+
+ PrintDebug("apic %u: (init_apic_state): Sizeof Interrupt Request Register %d, should be 32\n",
+ id, (uint_t)sizeof(apic->int_req_reg));
memset(apic->int_req_reg, 0, sizeof(apic->int_req_reg));
memset(apic->int_svc_reg, 0, sizeof(apic->int_svc_reg));
apic->tmr_init_cnt = 0x00000000;
apic->tmr_cur_cnt = 0x00000000;
- // TODO:
- // We need to figure out what the APIC ID is....
- apic->lapic_id.val = 0x00000000;
+ apic->lapic_id.val = id;
+
+ apic->icc_bus = icc;
// The P6 has 6 LVT entries, so we set the value to (6-1)...
apic->apic_ver.val = 0x80050010;
apic->ext_apic_feature.val = 0x00040007;
apic->ext_apic_ctrl.val = 0x00000000;
apic->spec_eoi.val = 0x00000000;
+
+ v3_lock_init(&(apic->lock));
}
-static int read_apic_msr(uint_t msr, v3_msr_t * dst, void * priv_data) {
+static int read_apic_msr(struct guest_info * core, uint_t msr, v3_msr_t * dst, void * priv_data) {
struct vm_device * dev = (struct vm_device *)priv_data;
- struct apic_state * apic = (struct apic_state *)dev->private_data;
+ struct apic_state * apics = (struct apic_state *)(dev->private_data);
+ struct apic_state * apic = &(apics[core->cpu_id]);
+
+ PrintDebug("apic %u: core %u: MSR read\n",apic->lapic_id.val,core->cpu_id);
+ v3_lock(apic->lock);
dst->value = apic->base_addr;
+ v3_unlock(apic->lock);
return 0;
}
-static int write_apic_msr(uint_t msr, v3_msr_t src, void * priv_data) {
+static int write_apic_msr(struct guest_info * core, uint_t msr, v3_msr_t src, void * priv_data) {
struct vm_device * dev = (struct vm_device *)priv_data;
- struct apic_state * apic = (struct apic_state *)dev->private_data;
- struct v3_shadow_region * old_reg = v3_get_shadow_region(dev->vm, apic->base_addr);
+ struct apic_state * apics = (struct apic_state *)(dev->private_data);
+ struct apic_state * apic = &(apics[core->cpu_id]);
+ struct v3_mem_region * old_reg = v3_get_mem_region(dev->vm, core->cpu_id, apic->base_addr);
+
+
+ PrintDebug("apic %u: core %u: MSR write\n",apic->lapic_id.val,core->cpu_id);
if (old_reg == NULL) {
// uh oh...
- PrintError("APIC Base address region does not exit...\n");
+ PrintError("apic %u: core %u: APIC Base address region does not exit...\n",apic->lapic_id.val,core->cpu_id);
return -1;
}
- v3_delete_shadow_region(dev->vm, old_reg);
+ v3_lock(apic->lock);
+
+ v3_delete_mem_region(dev->vm, old_reg);
apic->base_addr = src.value;
- if (v3_hook_full_mem(dev->vm, apic->base_addr, apic->base_addr + PAGE_SIZE_4KB, apic_read, apic_write, dev) == -1) {
- PrintError("Could not hook new APIC Base address\n");
+ if (v3_hook_full_mem(dev->vm, core->cpu_id, apic->base_addr, apic->base_addr + PAGE_SIZE_4KB, apic_read, apic_write, dev) == -1) {
+ PrintError("apic %u: core %u: Could not hook new APIC Base address\n",apic->lapic_id.val,core->cpu_id);
+ v3_unlock(apic->lock);
return -1;
}
+ v3_unlock(apic->lock);
return 0;
}
uchar_t * en_location = apic->int_en_reg + major_offset;
uchar_t flag = 0x1 << minor_offset;
+
+#if 1
+
if (irq_num <= 15) {
- PrintError("Attempting to raise an invalid interrupt: %d\n", irq_num);
+ PrintError("apic %u: core ?: Attempting to raise an invalid interrupt: %d\n", apic->lapic_id.val,irq_num);
return -1;
}
- PrintDebug("Raising APIC IRQ %d\n", irq_num);
+#endif
+
+
+ PrintDebug("apic %u: core ?: Raising APIC IRQ %d\n", apic->lapic_id.val,irq_num);
+
+ if (*req_location & flag) {
+ //V3_Print("Interrupts coallescing\n");
+ }
if (*en_location & flag) {
*req_location |= flag;
} else {
- PrintDebug("Interrupt not enabled... %.2x\n", *en_location);
+ PrintDebug("apic %u: core ?: Interrupt not enabled... %.2x\n", apic->lapic_id.val, *en_location);
return 0;
}
uchar_t flag = 0x1 << minor_offset;
uchar_t * svc_location = apic->int_svc_reg + major_offset;
- PrintDebug("Received APIC EOI for IRQ %d\n", isr_irq);
+ PrintDebug("apic %u: core ?: Received APIC EOI for IRQ %d\n", apic->lapic_id.val,isr_irq);
*svc_location &= ~flag;
if ((isr_irq == 238) ||
(isr_irq == 239)) {
- PrintError("Acking IRQ %d\n", isr_irq);
+ PrintError("apic %u: core ?: Acking IRQ %d\n", apic->lapic_id.val,isr_irq);
}
if (isr_irq == 238) {
}
#endif
} else {
- PrintError("Spurious EOI...\n");
+ //PrintError("apic %u: core ?: Spurious EOI...\n",apic->lapic_id.val);
}
return 0;
masked = apic->err_vec_tbl.mask;
break;
default:
- PrintError("Invalid APIC interrupt type\n");
+ PrintError("apic %u: core ?: Invalid APIC interrupt type\n",apic->lapic_id.val);
return -1;
}
// interrupt is masked, don't send
if (masked == 1) {
- PrintDebug("Inerrupt is masked\n");
+ PrintDebug("apic %u: core ?: Inerrupt is masked\n",apic->lapic_id.val);
return 0;
}
//PrintDebug("Activating internal APIC IRQ %d\n", vec_num);
return activate_apic_irq(apic, vec_num);
} else {
- PrintError("Unhandled Delivery Mode\n");
+ PrintError("apic %u: core ?: Unhandled Delivery Mode\n",apic->lapic_id.val);
return -1;
}
}
-static int apic_read(addr_t guest_addr, void * dst, uint_t length, void * priv_data) {
- struct vm_device * dev = (struct vm_device *)priv_data;
- struct apic_state * apic = (struct apic_state *)dev->private_data;
+static int apic_read(struct guest_info * core, addr_t guest_addr, void * dst, uint_t length, void * priv_data) {
+ struct apic_state * apic = (struct apic_state *)(priv_data);
addr_t reg_addr = guest_addr - apic->base_addr;
struct apic_msr * msr = (struct apic_msr *)&(apic->base_addr_msr.value);
uint32_t val = 0;
- PrintDebug("Read apic address space (%p)\n",
- (void *)guest_addr);
+ PrintDebug("apic %u: core %u: at %p: Read apic address space (%p)\n",apic->lapic_id.val,core->cpu_id, apic, (void *)guest_addr);
if (msr->apic_enable == 0) {
- PrintError("Write to APIC address space with disabled APIC\n");
+ PrintError("apic %u: core %u: Read from APIC address space with disabled APIC, apic msr=0x%llx\n",apic->lapic_id.val,core->cpu_id,apic->base_addr_msr.value);
+
return -1;
}
case SEOI_OFFSET:
default:
- PrintError("Read from Unhandled APIC Register: %x\n", (uint32_t)reg_addr);
- return -1;
+ PrintError("apic %u: core %u: Read from Unhandled APIC Register: %x (getting zero)\n", apic->lapic_id.val,core->cpu_id, (uint32_t)reg_addr);
+ // return -1;
+ val=0;
}
*val_ptr = val;
} else {
- PrintError("Invalid apic read length (%d)\n", length);
+ PrintError("apic %u: core %u: Invalid apic read length (%d)\n", apic->lapic_id.val,core->cpu_id, length);
return -1;
}
- PrintDebug("Read finished (val=%x)\n", *(uint32_t *)dst);
+ PrintDebug("apic %u: core %u: Read finished (val=%x)\n", apic->lapic_id.val,core->cpu_id, *(uint32_t *)dst);
return length;
}
-static int apic_write(addr_t guest_addr, void * src, uint_t length, void * priv_data) {
- struct vm_device * dev = (struct vm_device *)priv_data;
- struct apic_state * apic = (struct apic_state *)dev->private_data;
+/**
+ *
+ */
+static int apic_write(struct guest_info * core, addr_t guest_addr, void * src, uint_t length, void * priv_data) {
+ struct apic_state * apic = (struct apic_state *)(priv_data);
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;
- PrintDebug("Write to apic address space (%p) (val=%x)\n",
+ PrintDebug("apic %u: core %u: at %p and priv_data is at %p: Write to address space (%p) (val=%x)\n",
+ apic->lapic_id.val, core->cpu_id, apic,priv_data,
(void *)guest_addr, *(uint32_t *)src);
if (msr->apic_enable == 0) {
- PrintError("Write to APIC address space with disabled APIC\n");
+ PrintError("apic %u: core %u: Write to APIC address space with disabled APIC, apic msr=0x%llx\n",apic->lapic_id.val,core->cpu_id,apic->base_addr_msr.value);
return -1;
}
if (length != 4) {
- PrintError("Invalid apic write length (%d)\n", length);
+ PrintError("apic %u: core %u: Invalid apic write length (%d)\n", apic->lapic_id.val, length,core->cpu_id);
return -1;
}
case PPR_OFFSET:
case EXT_APIC_FEATURE_OFFSET:
#if 1
- PrintError("Attempting to write to read only register %p (ignored)\n", (void *)reg_addr);
+ PrintError("apic %u: core %u: Attempting to write to read only register %p (ignored)\n", apic->lapic_id.val,core->cpu_id, (void *)reg_addr);
#else
- PrintError("Attempting to write to read only register %p (error)\n", (void *)reg_addr);
+ PrintError("apic %u: core %u: Attempting to write to read only register %p (error)\n", apic->lapic_id.val,core->cpu_id, (void *)reg_addr);
return -1;
#endif
break;
// Data registers
case APIC_ID_OFFSET:
+ PrintDebug("apic %u: core %u: my id is being changed to %u\n",apic->lapic_id.val,core->cpu_id,op_val);
apic->lapic_id.val = op_val;
break;
case TPR_OFFSET:
break;
case INT_CMD_LO_OFFSET:
+ apic->int_cmd.lo = op_val;
+ // ICC???
+ PrintDebug("apic %u: core %u: sending cmd 0x%llx to apic %u\n",apic->lapic_id.val,core->cpu_id,
+ apic->int_cmd.val, apic->int_cmd.dst);
+ if (v3_icc_send_ipi(apic->icc_bus, apic->lapic_id.val, apic->int_cmd.val,apic->dst_fmt.val,0)==-1) {
+ return -1;
+ }
+ break;
case INT_CMD_HI_OFFSET:
+ apic->int_cmd.hi = op_val;
+ break;
// Unhandled Registers
case EXT_APIC_CMD_OFFSET:
case SEOI_OFFSET:
default:
- PrintError("Write to Unhandled APIC Register: %x\n", (uint32_t)reg_addr);
- return -1;
+ PrintError("apic %u: core %u: Write to Unhandled APIC Register: %x (ignored)\n", apic->lapic_id.val,core->cpu_id, (uint32_t)reg_addr);
+ // return -1;
}
- PrintDebug("Write finished\n");
+ PrintDebug("apic %u: core %u: Write finished\n",apic->lapic_id.val,core->cpu_id);
return length;
}
// returns 1 if an interrupt is pending, 0 otherwise
static int apic_intr_pending(struct guest_info * info, void * private_data) {
- struct vm_device * dev = (struct vm_device *)private_data;
- struct apic_state * apic = (struct apic_state *)dev->private_data;
+ struct apic_state * apic = (struct apic_state *)private_data;
int req_irq = get_highest_irr(apic);
int svc_irq = get_highest_isr(apic);
+ // PrintDebug("apic %u: core %u: req_irq=%d, svc_irq=%d\n",apic->lapic_id.val,info->cpu_id,req_irq,svc_irq);
+
if ((req_irq >= 0) &&
(req_irq > svc_irq)) {
return 1;
}
static int apic_get_intr_number(struct guest_info * info, void * private_data) {
- struct vm_device * dev = (struct vm_device *)private_data;
- struct apic_state * apic = (struct apic_state *)dev->private_data;
+ struct apic_state * apic = (struct apic_state *)private_data;
int req_irq = get_highest_irr(apic);
int svc_irq = get_highest_isr(apic);
return -1;
}
-static int apic_raise_intr(struct guest_info * info, void * private_data, int irq) {
-#ifdef CONFIG_CRAY_XT
- // The Seastar is connected directly to the LAPIC via LINT0 on the ICC bus
-
- if (irq == 238) {
- struct vm_device * dev = (struct vm_device *)private_data;
- struct apic_state * apic = (struct apic_state *)dev->private_data;
- return activate_apic_irq(apic, irq);
- }
-#endif
+static int apic_raise_intr(struct guest_info * info, int irq, void * private_data) {
+ struct apic_state * apic = (struct apic_state *)private_data;
- return 0;
+ return activate_apic_irq(apic, irq);
}
-static int apic_lower_intr(struct guest_info * info, void * private_data, int irq) {
- return 0;
-}
+
static int apic_begin_irq(struct guest_info * info, void * private_data, int irq) {
- struct vm_device * dev = (struct vm_device *)private_data;
- struct apic_state * apic = (struct apic_state *)dev->private_data;
+ struct apic_state * apic = (struct apic_state *)private_data;
int major_offset = (irq & ~0x00000007) >> 3;
int minor_offset = irq & 0x00000007;
uchar_t * req_location = apic->int_req_reg + major_offset;
uchar_t * svc_location = apic->int_svc_reg + major_offset;
uchar_t flag = 0x01 << minor_offset;
- *svc_location |= flag;
- *req_location &= ~flag;
-
-#ifdef CONFIG_CRAY_XT
- if ((irq == 238) || (irq == 239)) {
- PrintError("APIC: Begin IRQ %d (ISR=%x), (IRR=%x)\n", irq, *svc_location, *req_location);
+ if (*req_location & flag) {
+ // we will only pay attention to a begin irq if we
+ // know that we initiated it!
+ *svc_location |= flag;
+ *req_location &= ~flag;
+ } else {
+ // do nothing...
+ PrintDebug("apic %u: core %u: begin irq for %d ignored since I don't own it\n",
+ apic->lapic_id.val,info->cpu_id,irq);
}
-#endif
-
- return 0;
-}
-int v3_apic_raise_intr(struct guest_info * info, struct vm_device * apic_dev, int intr_num) {
- struct apic_state * apic = (struct apic_state *)apic_dev->private_data;
-
- if (activate_apic_irq(apic, intr_num) == -1) {
- PrintError("Error: Could not activate apic_irq\n");
- return -1;
- }
-
- v3_interrupt_cpu(info, 0);
-
return 0;
}
+
/* Timer Functions */
-static void apic_update_time(ullong_t cpu_cycles, ullong_t cpu_freq, void * priv_data) {
- struct vm_device * dev = (struct vm_device *)priv_data;
- struct apic_state * apic = (struct apic_state *)dev->private_data;
+static void apic_update_time(struct guest_info * info, ullong_t cpu_cycles, ullong_t cpu_freq, void * priv_data) {
+ struct apic_state * apic = (struct apic_state *)(priv_data);
// The 32 bit GCC runtime is a pile of shit
#ifdef __V3_64BIT__
uint64_t tmr_ticks = 0;
if ((apic->tmr_init_cnt == 0) ||
( (apic->tmr_vec_tbl.tmr_mode == APIC_TMR_ONESHOT) &&
(apic->tmr_cur_cnt == 0))) {
- //PrintDebug("APIC timer not yet initialized\n");
+ //PrintDebug("apic %u: core %u: APIC timer not yet initialized\n",apic->lapic_id.val,info->cpu_id);
return;
}
shift_num = 7;
break;
default:
- PrintError("Invalid Timer Divider configuration\n");
+ PrintError("apic %u: core %u: Invalid Timer Divider configuration\n",apic->lapic_id.val,info->cpu_id);
return;
}
apic->tmr_cur_cnt = 0;
// raise irq
- PrintDebug("Raising APIC Timer interrupt (periodic=%d) (icnt=%d) (div=%d)\n",
+ PrintDebug("apic %u: core %u: Raising APIC Timer interrupt (periodic=%d) (icnt=%d) (div=%d)\n", apic->lapic_id.val,info->cpu_id,
apic->tmr_vec_tbl.tmr_mode, apic->tmr_init_cnt, shift_num);
- if (apic_intr_pending(dev->vm, priv_data)) {
- PrintDebug("Overriding pending IRQ %d\n", apic_get_intr_number(dev->vm, priv_data));
+ if (apic_intr_pending(info, priv_data)) {
+ PrintDebug("apic %u: core %u: Overriding pending IRQ %d\n", apic->lapic_id.val,info->cpu_id, apic_get_intr_number(info, priv_data));
}
if (activate_internal_irq(apic, APIC_TMR_INT) == -1) {
- PrintError("Could not raise Timer interrupt\n");
+ PrintError("apic %u: core %u: Could not raise Timer interrupt\n",apic->lapic_id.val,info->cpu_id);
}
if (apic->tmr_vec_tbl.tmr_mode == APIC_TMR_PERIODIC) {
}
-
static struct intr_ctrl_ops intr_ops = {
.intr_pending = apic_intr_pending,
.get_intr_number = apic_get_intr_number,
- .raise_intr = apic_raise_intr,
.begin_irq = apic_begin_irq,
- .lower_intr = apic_lower_intr,
};
static int apic_free(struct vm_device * dev) {
- struct guest_info * info = dev->vm;
- v3_unhook_msr(info, BASE_ADDR_MSR);
+ /* TODO: This should crosscall to force an unhook on each CPU */
+
+ // struct apic_state * apic = (struct apic_state *)dev->private_data;
+
+ v3_unhook_msr(dev->vm, BASE_ADDR_MSR);
return 0;
}
-static int apic_init(struct guest_info * vm, void * cfg_data) {
- PrintDebug("Creating APIC\n");
+static int apic_should_deliver_flat(struct guest_info * core, uint8_t mda, void * private_data)
+{
+ struct apic_state * apic = (struct apic_state *)private_data;
+
+ if (mda==0xff || // broadcast or
+ (apic->log_dst.dst_log_id & mda)) { // I am in the set
+ return 1;
+ } else {
+ return 0;
+ }
+}
+
+static int apic_should_deliver_cluster(struct guest_info * core, uint8_t mda, void * private_data)
+{
+ struct apic_state * apic = (struct apic_state *)private_data;
+
+ if (mda==0xff || // broadcast or
+ ( ((mda & 0xf0) == (apic->log_dst.dst_log_id & 0xf0)) && // (I am in the cluster and
+ ((mda & 0x0f) & (apic->log_dst.dst_log_id & 0x0f)) ) ) { // I am in the set)
+ return 1;
+ } else {
+ return 0;
+ }
+}
+
+static struct v3_icc_ops icc_ops = {
+ .raise_intr = apic_raise_intr,
+ .should_deliver_flat = apic_should_deliver_flat,
+ .should_deliver_cluster = apic_should_deliver_cluster,
+};
+
+
+
+static int apic_init(struct v3_vm_info * vm, v3_cfg_tree_t * cfg) {
+ PrintDebug("apic: creating an APIC for each core\n");
+ char * name = v3_cfg_val(cfg, "name");
+ char * icc_name = v3_cfg_val(cfg,"bus");
+ struct vm_device * icc = v3_find_dev(vm, icc_name);
+ int i;
- struct apic_state * apic = (struct apic_state *)V3_Malloc(sizeof(struct apic_state));
+ if (!icc) {
+ PrintError("apic: Cannot find ICC Bus (%s)\n", icc_name);
+ return -1;
+ }
- struct vm_device * dev = v3_allocate_device("LAPIC", &dev_ops, apic);
+ // We allocate one apic per core
+ // APICs are accessed via index which correlates with the core's cpu_id
+ // 0..num_cores-1 at num_cores is the ioapic (one only)
+ struct apic_state * apic = (struct apic_state *)V3_Malloc(sizeof(struct apic_state) * vm->num_cores);
+
+ struct vm_device * dev = v3_allocate_device(name, &dev_ops, apic);
if (v3_attach_device(vm, dev) == -1) {
- PrintError("Could not attach device %s\n", "LAPIC");
+ PrintError("apic: Could not attach device %s\n", name);
return -1;
}
- v3_register_intr_controller(vm, &intr_ops, dev);
- v3_add_timer(vm, &timer_ops, dev);
+
+ for (i = 0; i < vm->num_cores; i++) {
+ struct guest_info * core = &(vm->cores[i]);
- init_apic_state(apic);
+ init_apic_state(&(apic[i]),i,icc);
- v3_hook_msr(vm, BASE_ADDR_MSR, read_apic_msr, write_apic_msr, dev);
+ v3_register_intr_controller(core, &intr_ops, &(apic[i]));
+
+ v3_add_timer(core, &timer_ops, &(apic[i]));
+
+ v3_hook_full_mem(vm, core->cpu_id, apic->base_addr, apic->base_addr + PAGE_SIZE_4KB, apic_read, apic_write, &(apic[i]));
+
+ v3_icc_register_apic(core, icc, i, &icc_ops, &(apic[i]));
+
+ PrintDebug("apic %u: (setup device): done, my id is %u\n",i,apic[i].lapic_id.val);
- v3_hook_full_mem(vm, apic->base_addr, apic->base_addr + PAGE_SIZE_4KB, apic_read, apic_write, dev);
+ }
+
+ for (i=0;i<vm->num_cores;i++) {
+ PrintDebug("apic: sanity check: apic %u (at %p) has id %u and msr value %llx\n",
+ i, &(apic[i]), apic[i].lapic_id.val, apic[i].base_addr_msr.value);
+ }
+ PrintDebug("apic: priv_data is at %p\n", apic);
+
+ v3_hook_msr(vm, BASE_ADDR_MSR, read_apic_msr, write_apic_msr, dev);
return 0;
}
