interrupts_and_traps.c

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

	 * bogus one in): if we fail here, the Guest will be killed. */
	if (!idt_present(lg->arch.idt[num].a, lg->arch.idt[num].b))
		return 0;
	set_guest_interrupt(lg, lg->arch.idt[num].a, lg->arch.idt[num].b,
			    has_err(num));
	return 1;
}

/*H:250 Here's the hard part: returning to the Host every time a trap happens
 * and then calling deliver_trap() and re-entering the Guest is slow.
 * Particularly because Guest userspace system calls are traps (usually trap
 * 128).
 *
 * So we'd like to set up the IDT to tell the CPU to deliver traps directly
 * into the Guest.  This is possible, but the complexities cause the size of
 * this file to double!  However, 150 lines of code is worth writing for taking
 * system calls down from 1750ns to 270ns.  Plus, if lguest didn't do it, all
 * the other hypervisors would beat it up at lunchtime.
 *
 * This routine indicates if a particular trap number could be delivered
 * directly. */
static int direct_trap(unsigned int num)
{
	/* Hardware interrupts don't go to the Guest at all (except system
	 * call). */
	if (num >= FIRST_EXTERNAL_VECTOR && !could_be_syscall(num))
		return 0;

	/* The Host needs to see page faults (for shadow paging and to save the
	 * fault address), general protection faults (in/out emulation) and
	 * device not available (TS handling), and of course, the hypercall
	 * trap. */
	return num != 14 && num != 13 && num != 7 && num != LGUEST_TRAP_ENTRY;
}
/*:*/

/*M:005 The Guest has the ability to turn its interrupt gates into trap gates,
 * if it is careful.  The Host will let trap gates can go directly to the
 * Guest, but the Guest needs the interrupts atomically disabled for an
 * interrupt gate.  It can do this by pointing the trap gate at instructions
 * within noirq_start and noirq_end, where it can safely disable interrupts. */

/*M:006 The Guests do not use the sysenter (fast system call) instruction,
 * because it's hardcoded to enter privilege level 0 and so can't go direct.
 * It's about twice as fast as the older "int 0x80" system call, so it might
 * still be worthwhile to handle it in the Switcher and lcall down to the
 * Guest.  The sysenter semantics are hairy tho: search for that keyword in
 * entry.S :*/

/*H:260 When we make traps go directly into the Guest, we need to make sure
 * the kernel stack is valid (ie. mapped in the page tables).  Otherwise, the
 * CPU trying to deliver the trap will fault while trying to push the interrupt
 * words on the stack: this is called a double fault, and it forces us to kill
 * the Guest.
 *
 * Which is deeply unfair, because (literally!) it wasn't the Guests' fault. */
void pin_stack_pages(struct lguest *lg)
{
	unsigned int i;

	/* Depending on the CONFIG_4KSTACKS option, the Guest can have one or
	 * two pages of stack space. */
	for (i = 0; i < lg->stack_pages; i++)
		/* The stack grows *upwards*, so the address we're given is the
		 * start of the page after the kernel stack.  Subtract one to
		 * get back onto the first stack page, and keep subtracting to
		 * get to the rest of the stack pages. */
		pin_page(lg, lg->esp1 - 1 - i * PAGE_SIZE);
}

/* Direct traps also mean that we need to know whenever the Guest wants to use
 * a different kernel stack, so we can change the IDT entries to use that
 * stack.  The IDT entries expect a virtual address, so unlike most addresses
 * the Guest gives us, the "esp" (stack pointer) value here is virtual, not
 * physical.
 *
 * In Linux each process has its own kernel stack, so this happens a lot: we
 * change stacks on each context switch. */
void guest_set_stack(struct lguest *lg, u32 seg, u32 esp, unsigned int pages)
{
	/* You are not allowed have a stack segment with privilege level 0: bad
	 * Guest! */
	if ((seg & 0x3) != GUEST_PL)
		kill_guest(lg, "bad stack segment %i", seg);
	/* We only expect one or two stack pages. */
	if (pages > 2)
		kill_guest(lg, "bad stack pages %u", pages);
	/* Save where the stack is, and how many pages */
	lg->ss1 = seg;
	lg->esp1 = esp;
	lg->stack_pages = pages;
	/* Make sure the new stack pages are mapped */
	pin_stack_pages(lg);
}

/* All this reference to mapping stacks leads us neatly into the other complex
 * part of the Host: page table handling. */

/*H:235 This is the routine which actually checks the Guest's IDT entry and
 * transfers it into the entry in "struct lguest": */
static void set_trap(struct lguest *lg, struct desc_struct *trap,
		     unsigned int num, u32 lo, u32 hi)
{
	u8 type = idt_type(lo, hi);

	/* We zero-out a not-present entry */
	if (!idt_present(lo, hi)) {
		trap->a = trap->b = 0;
		return;
	}

	/* We only support interrupt and trap gates. */
	if (type != 0xE && type != 0xF)
		kill_guest(lg, "bad IDT type %i", type);

	/* We only copy the handler address, present bit, privilege level and
	 * type.  The privilege level controls where the trap can be triggered
	 * manually with an "int" instruction.  This is usually GUEST_PL,
	 * except for system calls which userspace can use. */
	trap->a = ((__KERNEL_CS|GUEST_PL)<<16) | (lo&0x0000FFFF);
	trap->b = (hi&0xFFFFEF00);
}

/*H:230 While we're here, dealing with delivering traps and interrupts to the
 * Guest, we might as well complete the picture: how the Guest tells us where
 * it wants them to go.  This would be simple, except making traps fast
 * requires some tricks.
 *
 * We saw the Guest setting Interrupt Descriptor Table (IDT) entries with the
 * LHCALL_LOAD_IDT_ENTRY hypercall before: that comes here. */
void load_guest_idt_entry(struct lguest *lg, unsigned int num, u32 lo, u32 hi)
{
	/* Guest never handles: NMI, doublefault, spurious interrupt or
	 * hypercall.  We ignore when it tries to set them. */
	if (num == 2 || num == 8 || num == 15 || num == LGUEST_TRAP_ENTRY)
		return;

	/* Mark the IDT as changed: next time the Guest runs we'll know we have
	 * to copy this again. */
	lg->changed |= CHANGED_IDT;

	/* Check that the Guest doesn't try to step outside the bounds. */
	if (num >= ARRAY_SIZE(lg->arch.idt))
		kill_guest(lg, "Setting idt entry %u", num);
	else
		set_trap(lg, &lg->arch.idt[num], num, lo, hi);
}

/* The default entry for each interrupt points into the Switcher routines which
 * simply return to the Host.  The run_guest() loop will then call
 * deliver_trap() to bounce it back into the Guest. */
static void default_idt_entry(struct desc_struct *idt,
			      int trap,
			      const unsigned long handler)
{
	/* A present interrupt gate. */
	u32 flags = 0x8e00;

	/* Set the privilege level on the entry for the hypercall: this allows
	 * the Guest to use the "int" instruction to trigger it. */
	if (trap == LGUEST_TRAP_ENTRY)
		flags |= (GUEST_PL << 13);

	/* Now pack it into the IDT entry in its weird format. */
	idt->a = (LGUEST_CS<<16) | (handler&0x0000FFFF);
	idt->b = (handler&0xFFFF0000) | flags;
}

/* When the Guest first starts, we put default entries into the IDT. */
void setup_default_idt_entries(struct lguest_ro_state *state,
			       const unsigned long *def)
{
	unsigned int i;

	for (i = 0; i < ARRAY_SIZE(state->guest_idt); i++)
		default_idt_entry(&state->guest_idt[i], i, def[i]);
}

/*H:240 We don't use the IDT entries in the "struct lguest" directly, instead
 * we copy them into the IDT which we've set up for Guests on this CPU, just
 * before we run the Guest.  This routine does that copy. */
void copy_traps(const struct lguest *lg, struct desc_struct *idt,
		const unsigned long *def)
{
	unsigned int i;

	/* We can simply copy the direct traps, otherwise we use the default
	 * ones in the Switcher: they will return to the Host. */
	for (i = 0; i < ARRAY_SIZE(lg->arch.idt); i++) {
		/* If no Guest can ever override this trap, leave it alone. */
		if (!direct_trap(i))
			continue;

		/* Only trap gates (type 15) can go direct to the Guest.
		 * Interrupt gates (type 14) disable interrupts as they are
		 * entered, which we never let the Guest do.  Not present
		 * entries (type 0x0) also can't go direct, of course. */
		if (idt_type(lg->arch.idt[i].a, lg->arch.idt[i].b) == 0xF)
			idt[i] = lg->arch.idt[i];
		else
			/* Reset it to the default. */
			default_idt_entry(&idt[i], i, def[i]);
	}
}

/*H:200
 * The Guest Clock.
 *
 * There are two sources of virtual interrupts.  We saw one in lguest_user.c:
 * the Launcher sending interrupts for virtual devices.  The other is the Guest
 * timer interrupt.
 *
 * The Guest uses the LHCALL_SET_CLOCKEVENT hypercall to tell us how long to
 * the next timer interrupt (in nanoseconds).  We use the high-resolution timer
 * infrastructure to set a callback at that time.
 *
 * 0 means "turn off the clock". */
void guest_set_clockevent(struct lguest *lg, unsigned long delta)
{
	ktime_t expires;

	if (unlikely(delta == 0)) {
		/* Clock event device is shutting down. */
		hrtimer_cancel(&lg->hrt);
		return;
	}

	/* We use wallclock time here, so the Guest might not be running for
	 * all the time between now and the timer interrupt it asked for.  This
	 * is almost always the right thing to do. */
	expires = ktime_add_ns(ktime_get_real(), delta);
	hrtimer_start(&lg->hrt, expires, HRTIMER_MODE_ABS);
}

/* This is the function called when the Guest's timer expires. */
static enum hrtimer_restart clockdev_fn(struct hrtimer *timer)
{
	struct lguest *lg = container_of(timer, struct lguest, hrt);

	/* Remember the first interrupt is the timer interrupt. */
	set_bit(0, lg->irqs_pending);
	/* If the Guest is actually stopped, we need to wake it up. */
	if (lg->halted)
		wake_up_process(lg->tsk);
	return HRTIMER_NORESTART;
}

/* This sets up the timer for this Guest. */
void init_clockdev(struct lguest *lg)
{
	hrtimer_init(&lg->hrt, CLOCK_REALTIME, HRTIMER_MODE_ABS);
	lg->hrt.function = clockdev_fn;
}