interrupts_and_traps.c

来自「linux 内核源代码」· C语言 代码 · 共 509 行 · 第 1/2 页

/*P:800 Interrupts (traps) are complicated enough to earn their own file.
 * There are three classes of interrupts:
 *
 * 1) Real hardware interrupts which occur while we're running the Guest,
 * 2) Interrupts for virtual devices attached to the Guest, and
 * 3) Traps and faults from the Guest.
 *
 * Real hardware interrupts must be delivered to the Host, not the Guest.
 * Virtual interrupts must be delivered to the Guest, but we make them look
 * just like real hardware would deliver them.  Traps from the Guest can be set
 * up to go directly back into the Guest, but sometimes the Host wants to see
 * them first, so we also have a way of "reflecting" them into the Guest as if
 * they had been delivered to it directly. :*/
#include <linux/uaccess.h>
#include <linux/interrupt.h>
#include <linux/module.h>
#include "lg.h"

/* Allow Guests to use a non-128 (ie. non-Linux) syscall trap. */
static unsigned int syscall_vector = SYSCALL_VECTOR;
module_param(syscall_vector, uint, 0444);

/* The address of the interrupt handler is split into two bits: */
static unsigned long idt_address(u32 lo, u32 hi)
{
	return (lo & 0x0000FFFF) | (hi & 0xFFFF0000);
}

/* The "type" of the interrupt handler is a 4 bit field: we only support a
 * couple of types. */
static int idt_type(u32 lo, u32 hi)
{
	return (hi >> 8) & 0xF;
}

/* An IDT entry can't be used unless the "present" bit is set. */
static int idt_present(u32 lo, u32 hi)
{
	return (hi & 0x8000);
}

/* We need a helper to "push" a value onto the Guest's stack, since that's a
 * big part of what delivering an interrupt does. */
static void push_guest_stack(struct lguest *lg, unsigned long *gstack, u32 val)
{
	/* Stack grows upwards: move stack then write value. */
	*gstack -= 4;
	lgwrite(lg, *gstack, u32, val);
}

/*H:210 The set_guest_interrupt() routine actually delivers the interrupt or
 * trap.  The mechanics of delivering traps and interrupts to the Guest are the
 * same, except some traps have an "error code" which gets pushed onto the
 * stack as well: the caller tells us if this is one.
 *
 * "lo" and "hi" are the two parts of the Interrupt Descriptor Table for this
 * interrupt or trap.  It's split into two parts for traditional reasons: gcc
 * on i386 used to be frightened by 64 bit numbers.
 *
 * We set up the stack just like the CPU does for a real interrupt, so it's
 * identical for the Guest (and the standard "iret" instruction will undo
 * it). */
static void set_guest_interrupt(struct lguest *lg, u32 lo, u32 hi, int has_err)
{
	unsigned long gstack, origstack;
	u32 eflags, ss, irq_enable;
	unsigned long virtstack;

	/* There are two cases for interrupts: one where the Guest is already
	 * in the kernel, and a more complex one where the Guest is in
	 * userspace.  We check the privilege level to find out. */
	if ((lg->regs->ss&0x3) != GUEST_PL) {
		/* The Guest told us their kernel stack with the SET_STACK
		 * hypercall: both the virtual address and the segment */
		virtstack = lg->esp1;
		ss = lg->ss1;

		origstack = gstack = guest_pa(lg, virtstack);
		/* We push the old stack segment and pointer onto the new
		 * stack: when the Guest does an "iret" back from the interrupt
		 * handler the CPU will notice they're dropping privilege
		 * levels and expect these here. */
		push_guest_stack(lg, &gstack, lg->regs->ss);
		push_guest_stack(lg, &gstack, lg->regs->esp);
	} else {
		/* We're staying on the same Guest (kernel) stack. */
		virtstack = lg->regs->esp;
		ss = lg->regs->ss;

		origstack = gstack = guest_pa(lg, virtstack);
	}

	/* Remember that we never let the Guest actually disable interrupts, so
	 * the "Interrupt Flag" bit is always set.  We copy that bit from the
	 * Guest's "irq_enabled" field into the eflags word: we saw the Guest
	 * copy it back in "lguest_iret". */
	eflags = lg->regs->eflags;
	if (get_user(irq_enable, &lg->lguest_data->irq_enabled) == 0
	    && !(irq_enable & X86_EFLAGS_IF))
		eflags &= ~X86_EFLAGS_IF;

	/* An interrupt is expected to push three things on the stack: the old
	 * "eflags" word, the old code segment, and the old instruction
	 * pointer. */
	push_guest_stack(lg, &gstack, eflags);
	push_guest_stack(lg, &gstack, lg->regs->cs);
	push_guest_stack(lg, &gstack, lg->regs->eip);

	/* For the six traps which supply an error code, we push that, too. */
	if (has_err)
		push_guest_stack(lg, &gstack, lg->regs->errcode);

	/* Now we've pushed all the old state, we change the stack, the code
	 * segment and the address to execute. */
	lg->regs->ss = ss;
	lg->regs->esp = virtstack + (gstack - origstack);
	lg->regs->cs = (__KERNEL_CS|GUEST_PL);
	lg->regs->eip = idt_address(lo, hi);

	/* There are two kinds of interrupt handlers: 0xE is an "interrupt
	 * gate" which expects interrupts to be disabled on entry. */
	if (idt_type(lo, hi) == 0xE)
		if (put_user(0, &lg->lguest_data->irq_enabled))
			kill_guest(lg, "Disabling interrupts");
}

/*H:205
 * Virtual Interrupts.
 *
 * maybe_do_interrupt() gets called before every entry to the Guest, to see if
 * we should divert the Guest to running an interrupt handler. */
void maybe_do_interrupt(struct lguest *lg)
{
	unsigned int irq;
	DECLARE_BITMAP(blk, LGUEST_IRQS);
	struct desc_struct *idt;

	/* If the Guest hasn't even initialized yet, we can do nothing. */
	if (!lg->lguest_data)
		return;

	/* Take our "irqs_pending" array and remove any interrupts the Guest
	 * wants blocked: the result ends up in "blk". */
	if (copy_from_user(&blk, lg->lguest_data->blocked_interrupts,
			   sizeof(blk)))
		return;

	bitmap_andnot(blk, lg->irqs_pending, blk, LGUEST_IRQS);

	/* Find the first interrupt. */
	irq = find_first_bit(blk, LGUEST_IRQS);
	/* None?  Nothing to do */
	if (irq >= LGUEST_IRQS)
		return;

	/* They may be in the middle of an iret, where they asked us never to
	 * deliver interrupts. */
	if (lg->regs->eip >= lg->noirq_start && lg->regs->eip < lg->noirq_end)
		return;

	/* If they're halted, interrupts restart them. */
	if (lg->halted) {
		/* Re-enable interrupts. */
		if (put_user(X86_EFLAGS_IF, &lg->lguest_data->irq_enabled))
			kill_guest(lg, "Re-enabling interrupts");
		lg->halted = 0;
	} else {
		/* Otherwise we check if they have interrupts disabled. */
		u32 irq_enabled;
		if (get_user(irq_enabled, &lg->lguest_data->irq_enabled))
			irq_enabled = 0;
		if (!irq_enabled)
			return;
	}

	/* Look at the IDT entry the Guest gave us for this interrupt.  The
	 * first 32 (FIRST_EXTERNAL_VECTOR) entries are for traps, so we skip
	 * over them. */
	idt = &lg->arch.idt[FIRST_EXTERNAL_VECTOR+irq];
	/* If they don't have a handler (yet?), we just ignore it */
	if (idt_present(idt->a, idt->b)) {
		/* OK, mark it no longer pending and deliver it. */
		clear_bit(irq, lg->irqs_pending);
		/* set_guest_interrupt() takes the interrupt descriptor and a
		 * flag to say whether this interrupt pushes an error code onto
		 * the stack as well: virtual interrupts never do. */
		set_guest_interrupt(lg, idt->a, idt->b, 0);
	}

	/* Every time we deliver an interrupt, we update the timestamp in the
	 * Guest's lguest_data struct.  It would be better for the Guest if we
	 * did this more often, but it can actually be quite slow: doing it
	 * here is a compromise which means at least it gets updated every
	 * timer interrupt. */
	write_timestamp(lg);
}
/*:*/

/* Linux uses trap 128 for system calls.  Plan9 uses 64, and Ron Minnich sent
 * me a patch, so we support that too.  It'd be a big step for lguest if half
 * the Plan 9 user base were to start using it.
 *
 * Actually now I think of it, it's possible that Ron *is* half the Plan 9
 * userbase.  Oh well. */
static bool could_be_syscall(unsigned int num)
{
	/* Normal Linux SYSCALL_VECTOR or reserved vector? */
	return num == SYSCALL_VECTOR || num == syscall_vector;
}

/* The syscall vector it wants must be unused by Host. */
bool check_syscall_vector(struct lguest *lg)
{
	u32 vector;

	if (get_user(vector, &lg->lguest_data->syscall_vec))
		return false;

	return could_be_syscall(vector);
}

int init_interrupts(void)
{
	/* If they want some strange system call vector, reserve it now */
	if (syscall_vector != SYSCALL_VECTOR
	    && test_and_set_bit(syscall_vector, used_vectors)) {
		printk("lg: couldn't reserve syscall %u\n", syscall_vector);
		return -EBUSY;
	}
	return 0;
}

void free_interrupts(void)
{
	if (syscall_vector != SYSCALL_VECTOR)
		clear_bit(syscall_vector, used_vectors);
}

/*H:220 Now we've got the routines to deliver interrupts, delivering traps
 * like page fault is easy.  The only trick is that Intel decided that some
 * traps should have error codes: */
static int has_err(unsigned int trap)
{
	return (trap == 8 || (trap >= 10 && trap <= 14) || trap == 17);
}

/* deliver_trap() returns true if it could deliver the trap. */
int deliver_trap(struct lguest *lg, unsigned int num)
{
	/* Trap numbers are always 8 bit, but we set an impossible trap number
	 * for traps inside the Switcher, so check that here. */
	if (num >= ARRAY_SIZE(lg->arch.idt))
		return 0;

	/* Early on the Guest hasn't set the IDT entries (or maybe it put a