acrn-hypervisor/hypervisor/arch/x86/irq.c

500 lines
12 KiB
C

/*
* Copyright (C) 2018 Intel Corporation. All rights reserved.
*
* SPDX-License-Identifier: BSD-3-Clause
*/
#include <types.h>
#include <errno.h>
#include <bits.h>
#include <spinlock.h>
#include <per_cpu.h>
#include <io.h>
#include <irq.h>
#include <idt.h>
#include <ioapic.h>
#include <lapic.h>
#include <softirq.h>
#include <vboot.h>
#include <dump.h>
#include <logmsg.h>
#include <vmx.h>
static spinlock_t exception_spinlock = { .head = 0U, .tail = 0U, };
static spinlock_t irq_alloc_spinlock = { .head = 0U, .tail = 0U, };
uint64_t irq_alloc_bitmap[IRQ_ALLOC_BITMAP_SIZE];
struct irq_desc irq_desc_array[NR_IRQS];
static uint32_t vector_to_irq[NR_MAX_VECTOR + 1];
spurious_handler_t spurious_handler;
struct static_mapping_table {
uint32_t irq;
uint32_t vector;
};
static struct static_mapping_table irq_static_mappings[NR_STATIC_MAPPINGS] = {
{TIMER_IRQ, TIMER_VECTOR},
{NOTIFY_VCPU_IRQ, NOTIFY_VCPU_VECTOR},
{POSTED_INTR_IRQ, POSTED_INTR_VECTOR},
{PMI_IRQ, PMI_VECTOR},
};
/*
* alloc an free irq if req_irq is IRQ_INVALID, or else set assigned
* return: irq num on success, IRQ_INVALID on failure
*/
uint32_t alloc_irq_num(uint32_t req_irq)
{
uint32_t irq = req_irq;
uint64_t rflags;
uint32_t ret;
if ((irq >= NR_IRQS) && (irq != IRQ_INVALID)) {
pr_err("[%s] invalid req_irq %u", __func__, req_irq);
ret = IRQ_INVALID;
} else {
spinlock_irqsave_obtain(&irq_alloc_spinlock, &rflags);
if (irq == IRQ_INVALID) {
/* if no valid irq num given, find a free one */
irq = (uint32_t)ffz64_ex(irq_alloc_bitmap, NR_IRQS);
}
if (irq >= NR_IRQS) {
irq = IRQ_INVALID;
} else {
bitmap_set_nolock((uint16_t)(irq & 0x3FU),
irq_alloc_bitmap + (irq >> 6U));
}
spinlock_irqrestore_release(&irq_alloc_spinlock, rflags);
ret = irq;
}
return ret;
}
/*
* @pre: irq is not in irq_static_mappings
* free irq num allocated via alloc_irq_num()
*/
static void free_irq_num(uint32_t irq)
{
uint64_t rflags;
if (irq < NR_IRQS) {
if (!is_ioapic_irq(irq)) {
spinlock_irqsave_obtain(&irq_alloc_spinlock, &rflags);
(void)bitmap_test_and_clear_nolock((uint16_t)(irq & 0x3FU),
irq_alloc_bitmap + (irq >> 6U));
spinlock_irqrestore_release(&irq_alloc_spinlock, rflags);
}
}
}
/*
* alloc an vectror and bind it to irq
* for legacy_irq (irq num < 16) and static mapped ones, do nothing
* if mapping is correct.
* retval: valid vector num on susccess, VECTOR_INVALID on failure.
*/
uint32_t alloc_irq_vector(uint32_t irq)
{
uint32_t vr;
struct irq_desc *desc;
uint64_t rflags;
uint32_t ret;
if (irq < NR_IRQS) {
desc = &irq_desc_array[irq];
if (desc->vector != VECTOR_INVALID) {
if (vector_to_irq[desc->vector] == irq) {
/* statically binded */
vr = desc->vector;
} else {
pr_err("[%s] irq[%u]:vector[%u] mismatch",
__func__, irq, desc->vector);
vr = VECTOR_INVALID;
}
} else {
/* alloc a vector between:
* VECTOR_DYNAMIC_START ~ VECTOR_DYNAMC_END
*/
spinlock_irqsave_obtain(&irq_alloc_spinlock, &rflags);
for (vr = VECTOR_DYNAMIC_START;
vr <= VECTOR_DYNAMIC_END; vr++) {
if (vector_to_irq[vr] == IRQ_INVALID) {
desc->vector = vr;
vector_to_irq[vr] = irq;
break;
}
}
vr = (vr > VECTOR_DYNAMIC_END) ? VECTOR_INVALID : vr;
spinlock_irqrestore_release(&irq_alloc_spinlock, rflags);
}
ret = vr;
} else {
pr_err("invalid irq[%u] to alloc vector", irq);
ret = VECTOR_INVALID;
}
return ret;
}
/* free the vector allocated via alloc_irq_vector() */
static void free_irq_vector(uint32_t irq)
{
struct irq_desc *desc;
uint32_t vr;
uint64_t rflags;
if (irq < NR_IRQS) {
desc = &irq_desc_array[irq];
if ((irq >= NR_LEGACY_IRQ) && (desc->vector < VECTOR_FIXED_START)) {
/* do nothing for LEGACY_IRQ and static allocated ones */
spinlock_irqsave_obtain(&irq_alloc_spinlock, &rflags);
vr = desc->vector;
desc->vector = VECTOR_INVALID;
vr &= NR_MAX_VECTOR;
if (vector_to_irq[vr] == irq) {
vector_to_irq[vr] = IRQ_INVALID;
}
spinlock_irqrestore_release(&irq_alloc_spinlock, rflags);
}
}
}
/*
* There are four cases as to irq/vector allocation:
* case 1: req_irq = IRQ_INVALID
* caller did not know which irq to use, and want system to
* allocate available irq for it. These irq are in range:
* nr_gsi ~ NR_IRQS
* an irq will be allocated and a vector will be assigned to this
* irq automatically.
* case 2: req_irq >= NR_LAGACY_IRQ and irq < nr_gsi
* caller want to add device ISR handler into ioapic pins.
* a vector will automatically assigned.
* case 3: req_irq >=0 and req_irq < NR_LEGACY_IRQ
* caller want to add device ISR handler into ioapic pins, which
* is a legacy irq, vector already reserved.
* Nothing to do in this case.
* case 4: irq with speical type (not from IOAPIC/MSI)
* These irq value are pre-defined for Timer, IPI, Spurious etc,
* which is listed in irq_static_mappings[].
* Nothing to do in this case.
*
* return value: valid irq (>=0) on success, otherwise errno (< 0).
*/
int32_t request_irq(uint32_t req_irq, irq_action_t action_fn, void *priv_data,
uint32_t flags)
{
struct irq_desc *desc;
uint32_t irq, vector;
uint64_t rflags;
int32_t ret;
irq = alloc_irq_num(req_irq);
if (irq == IRQ_INVALID) {
pr_err("[%s] invalid irq num", __func__);
ret = -EINVAL;
} else {
vector = alloc_irq_vector(irq);
if (vector == VECTOR_INVALID) {
pr_err("[%s] failed to alloc vector for irq %u",
__func__, irq);
free_irq_num(irq);
ret = -EINVAL;
} else {
desc = &irq_desc_array[irq];
spinlock_irqsave_obtain(&desc->lock, &rflags);
if (desc->action == NULL) {
desc->flags = flags;
desc->priv_data = priv_data;
desc->action = action_fn;
spinlock_irqrestore_release(&desc->lock, rflags);
ret = (int32_t)irq;
dev_dbg(DBG_LEVEL_IRQ, "[%s] irq%d vr:0x%x", __func__, irq, desc->vector);
} else {
spinlock_irqrestore_release(&desc->lock, rflags);
ret = -EBUSY;
pr_err("%s: request irq(%u) vr(%u) failed, already requested", __func__,
irq, irq_to_vector(irq));
}
}
}
return ret;
}
void free_irq(uint32_t irq)
{
uint64_t rflags;
struct irq_desc *desc;
if (irq < NR_IRQS) {
desc = &irq_desc_array[irq];
dev_dbg(DBG_LEVEL_IRQ, "[%s] irq%d vr:0x%x",
__func__, irq, irq_to_vector(irq));
free_irq_vector(irq);
free_irq_num(irq);
spinlock_irqsave_obtain(&desc->lock, &rflags);
desc->action = NULL;
desc->priv_data = NULL;
desc->flags = IRQF_NONE;
spinlock_irqrestore_release(&desc->lock, rflags);
}
}
void set_irq_trigger_mode(uint32_t irq, bool is_level_triggered)
{
uint64_t rflags;
struct irq_desc *desc;
if (irq < NR_IRQS) {
desc = &irq_desc_array[irq];
spinlock_irqsave_obtain(&desc->lock, &rflags);
if (is_level_triggered == true) {
desc->flags |= IRQF_LEVEL;
} else {
desc->flags &= ~IRQF_LEVEL;
}
spinlock_irqrestore_release(&desc->lock, rflags);
}
}
uint32_t irq_to_vector(uint32_t irq)
{
uint32_t ret;
if (irq < NR_IRQS) {
ret = irq_desc_array[irq].vector;
} else {
ret = VECTOR_INVALID;
}
return ret;
}
static void handle_spurious_interrupt(uint32_t vector)
{
send_lapic_eoi();
get_cpu_var(spurious)++;
pr_warn("Spurious vector: 0x%x.", vector);
if (spurious_handler != NULL) {
spurious_handler(vector);
}
}
static inline bool irq_need_mask(const struct irq_desc *desc)
{
/* level triggered gsi should be masked */
return (((desc->flags & IRQF_LEVEL) != 0U)
&& is_ioapic_irq(desc->irq));
}
static inline bool irq_need_unmask(const struct irq_desc *desc)
{
/* level triggered gsi for non-ptdev should be unmasked */
return (((desc->flags & IRQF_LEVEL) != 0U)
&& ((desc->flags & IRQF_PT) == 0U)
&& is_ioapic_irq(desc->irq));
}
static inline void handle_irq(const struct irq_desc *desc)
{
irq_action_t action = desc->action;
if (irq_need_mask(desc)) {
ioapic_gsi_mask_irq(desc->irq);
}
/* Send EOI to LAPIC/IOAPIC IRR */
send_lapic_eoi();
if (action != NULL) {
action(desc->irq, desc->priv_data);
}
if (irq_need_unmask(desc)) {
ioapic_gsi_unmask_irq(desc->irq);
}
}
/* do_IRQ() */
void dispatch_interrupt(const struct intr_excp_ctx *ctx)
{
uint32_t vr = ctx->vector;
uint32_t irq = vector_to_irq[vr];
struct irq_desc *desc;
/* The value from vector_to_irq[] must be:
* IRQ_INVALID, which means the vector is not allocated;
* or
* < NR_IRQS, which is the irq number it bound with;
* Any other value means there is something wrong.
*/
if (irq < NR_IRQS) {
desc = &irq_desc_array[irq];
per_cpu(irq_count, get_pcpu_id())[irq]++;
if ((vr == desc->vector) &&
bitmap_test((uint16_t)(irq & 0x3FU), irq_alloc_bitmap + (irq >> 6U))) {
#ifdef PROFILING_ON
/* Saves ctx info into irq_desc */
desc->ctx_rip = ctx->rip;
desc->ctx_rflags = ctx->rflags;
desc->ctx_cs = ctx->cs;
#endif
handle_irq(desc);
}
} else {
handle_spurious_interrupt(vr);
}
do_softirq();
}
void dispatch_exception(struct intr_excp_ctx *ctx)
{
uint16_t pcpu_id = get_pcpu_id();
/* Obtain lock to ensure exception dump doesn't get corrupted */
spinlock_obtain(&exception_spinlock);
/* Dump exception context */
dump_exception(ctx, pcpu_id);
/* Release lock to let other CPUs handle exception */
spinlock_release(&exception_spinlock);
/* Halt the CPU */
cpu_dead();
}
void handle_nmi(__unused struct intr_excp_ctx *ctx)
{
uint32_t value32;
/*
* There is a window where we may miss the current request in this
* notification period when the work flow is as the following:
*
* CPUx + + CPUr
* | |
* | +--+
* | | | Handle pending req
* | <--+
* +--+ |
* | | Set req flag |
* <--+ |
* +------------------>---+
* | Send NMI | | Handle NMI
* | <--+
* | |
* | |
* | +--> vCPU enter
* | |
* + +
*
* So, here we enable the NMI-window exiting to trigger the next vmexit
* once there is no "virtual-NMI blocking" after vCPU enter into VMX non-root
* mode. Then we can process the pending request on time.
*/
value32 = exec_vmread32(VMX_PROC_VM_EXEC_CONTROLS);
value32 |= VMX_PROCBASED_CTLS_NMI_WINEXIT;
exec_vmwrite32(VMX_PROC_VM_EXEC_CONTROLS, value32);
}
static void init_irq_descs(void)
{
uint32_t i;
for (i = 0U; i < NR_IRQS; i++) {
irq_desc_array[i].irq = i;
irq_desc_array[i].vector = VECTOR_INVALID;
spinlock_init(&irq_desc_array[i].lock);
}
for (i = 0U; i <= NR_MAX_VECTOR; i++) {
vector_to_irq[i] = IRQ_INVALID;
}
/* init fixed mapping for specific irq and vector */
for (i = 0U; i < NR_STATIC_MAPPINGS; i++) {
uint32_t irq = irq_static_mappings[i].irq;
uint32_t vr = irq_static_mappings[i].vector;
irq_desc_array[irq].vector = vr;
vector_to_irq[vr] = irq;
bitmap_set_nolock((uint16_t)(irq & 0x3FU),
irq_alloc_bitmap + (irq >> 6U));
}
}
static void disable_pic_irqs(void)
{
pio_write8(0xffU, 0xA1U);
pio_write8(0xffU, 0x21U);
}
void init_default_irqs(uint16_t cpu_id)
{
if (cpu_id == BSP_CPU_ID) {
init_irq_descs();
/* we use ioapic only, disable legacy PIC */
disable_pic_irqs();
ioapic_setup_irqs();
init_softirq();
}
}
static inline void fixup_idt(const struct host_idt_descriptor *idtd)
{
uint32_t i;
union idt_64_descriptor *idt_desc = (union idt_64_descriptor *)idtd->idt;
uint32_t entry_hi_32, entry_lo_32;
for (i = 0U; i < HOST_IDT_ENTRIES; i++) {
entry_lo_32 = idt_desc[i].fields.offset_63_32;
entry_hi_32 = idt_desc[i].fields.rsvd;
idt_desc[i].fields.rsvd = 0U;
idt_desc[i].fields.offset_63_32 = entry_hi_32;
idt_desc[i].fields.high32.bits.offset_31_16 = entry_lo_32 >> 16U;
idt_desc[i].fields.low32.bits.offset_15_0 = entry_lo_32 & 0xffffUL;
}
}
static inline void set_idt(struct host_idt_descriptor *idtd)
{
asm volatile (" lidtq %[idtd]\n" : /* no output parameters */
: /* input parameters */
[idtd] "m"(*idtd));
}
void init_interrupt(uint16_t pcpu_id)
{
struct host_idt_descriptor *idtd = &HOST_IDTR;
if (pcpu_id == BSP_CPU_ID) {
fixup_idt(idtd);
}
set_idt(idtd);
init_lapic(pcpu_id);
init_default_irqs(pcpu_id);
init_vboot_irq();
}