-/* Actual hypercalls, which allow guests to actually do something.
- Copyright (C) 2006 Rusty Russell IBM Corporation
+/*P:500 Just as userspace programs request kernel operations through a system
+ * call, the Guest requests Host operations through a "hypercall". You might
+ * notice this nomenclature doesn't really follow any logic, but the name has
+ * been around for long enough that we're stuck with it. As you'd expect, this
+ * code is basically a one big switch statement. :*/
+
+/* Copyright (C) 2006 Rusty Russell IBM Corporation
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
#include <linux/mm.h>
#include <asm/page.h>
#include <asm/pgtable.h>
-#include <irq_vectors.h>
#include "lg.h"
-static void do_hcall(struct lguest *lg, struct lguest_regs *regs)
+/*H:120 This is the core hypercall routine: where the Guest gets what it wants.
+ * Or gets killed. Or, in the case of LHCALL_CRASH, both. */
+static void do_hcall(struct lguest *lg, struct hcall_args *args)
{
- switch (regs->eax) {
+ switch (args->arg0) {
case LHCALL_FLUSH_ASYNC:
+ /* This call does nothing, except by breaking out of the Guest
+ * it makes us process all the asynchronous hypercalls. */
break;
case LHCALL_LGUEST_INIT:
+ /* You can't get here unless you're already initialized. Don't
+ * do that. */
kill_guest(lg, "already have lguest_data");
break;
- case LHCALL_CRASH: {
+ case LHCALL_SHUTDOWN: {
+ /* Shutdown is such a trivial hypercall that we do it in four
+ * lines right here. */
char msg[128];
- lgread(lg, msg, regs->edx, sizeof(msg));
+ /* If the lgread fails, it will call kill_guest() itself; the
+ * kill_guest() with the message will be ignored. */
+ __lgread(lg, msg, args->arg1, sizeof(msg));
msg[sizeof(msg)-1] = '\0';
kill_guest(lg, "CRASH: %s", msg);
+ if (args->arg2 == LGUEST_SHUTDOWN_RESTART)
+ lg->dead = ERR_PTR(-ERESTART);
break;
}
case LHCALL_FLUSH_TLB:
- if (regs->edx)
+ /* FLUSH_TLB comes in two flavors, depending on the
+ * argument: */
+ if (args->arg1)
guest_pagetable_clear_all(lg);
else
guest_pagetable_flush_user(lg);
break;
- case LHCALL_GET_WALLCLOCK: {
- struct timespec ts;
- ktime_get_real_ts(&ts);
- regs->eax = ts.tv_sec;
- break;
- }
- case LHCALL_BIND_DMA:
- regs->eax = bind_dma(lg, regs->edx, regs->ebx,
- regs->ecx >> 8, regs->ecx & 0xFF);
- break;
- case LHCALL_SEND_DMA:
- send_dma(lg, regs->edx, regs->ebx);
- break;
- case LHCALL_LOAD_GDT:
- load_guest_gdt(lg, regs->edx, regs->ebx);
- break;
- case LHCALL_LOAD_IDT_ENTRY:
- load_guest_idt_entry(lg, regs->edx, regs->ebx, regs->ecx);
- break;
+
+ /* All these calls simply pass the arguments through to the right
+ * routines. */
case LHCALL_NEW_PGTABLE:
- guest_new_pagetable(lg, regs->edx);
+ guest_new_pagetable(lg, args->arg1);
break;
case LHCALL_SET_STACK:
- guest_set_stack(lg, regs->edx, regs->ebx, regs->ecx);
+ guest_set_stack(lg, args->arg1, args->arg2, args->arg3);
break;
case LHCALL_SET_PTE:
- guest_set_pte(lg, regs->edx, regs->ebx, mkgpte(regs->ecx));
+ guest_set_pte(lg, args->arg1, args->arg2, __pte(args->arg3));
break;
case LHCALL_SET_PMD:
- guest_set_pmd(lg, regs->edx, regs->ebx);
- break;
- case LHCALL_LOAD_TLS:
- guest_load_tls(lg, regs->edx);
+ guest_set_pmd(lg, args->arg1, args->arg2);
break;
case LHCALL_SET_CLOCKEVENT:
- guest_set_clockevent(lg, regs->edx);
+ guest_set_clockevent(lg, args->arg1);
break;
case LHCALL_TS:
- lg->ts = regs->edx;
+ /* This sets the TS flag, as we saw used in run_guest(). */
+ lg->ts = args->arg1;
break;
case LHCALL_HALT:
+ /* Similarly, this sets the halted flag for run_guest(). */
lg->halted = 1;
break;
+ case LHCALL_NOTIFY:
+ lg->pending_notify = args->arg1;
+ break;
default:
- kill_guest(lg, "Bad hypercall %li\n", regs->eax);
+ /* It should be an architecture-specific hypercall. */
+ if (lguest_arch_do_hcall(lg, args))
+ kill_guest(lg, "Bad hypercall %li\n", args->arg0);
}
}
-
-/* We always do queued calls before actual hypercall. */
+/*:*/
+
+/*H:124 Asynchronous hypercalls are easy: we just look in the array in the
+ * Guest's "struct lguest_data" to see if any new ones are marked "ready".
+ *
+ * We are careful to do these in order: obviously we respect the order the
+ * Guest put them in the ring, but we also promise the Guest that they will
+ * happen before any normal hypercall (which is why we check this before
+ * checking for a normal hcall). */
static void do_async_hcalls(struct lguest *lg)
{
unsigned int i;
u8 st[LHCALL_RING_SIZE];
+ /* For simplicity, we copy the entire call status array in at once. */
if (copy_from_user(&st, &lg->lguest_data->hcall_status, sizeof(st)))
return;
+ /* We process "struct lguest_data"s hcalls[] ring once. */
for (i = 0; i < ARRAY_SIZE(st); i++) {
- struct lguest_regs regs;
+ struct hcall_args args;
+ /* We remember where we were up to from last time. This makes
+ * sure that the hypercalls are done in the order the Guest
+ * places them in the ring. */
unsigned int n = lg->next_hcall;
+ /* 0xFF means there's no call here (yet). */
if (st[n] == 0xFF)
break;
+ /* OK, we have hypercall. Increment the "next_hcall" cursor,
+ * and wrap back to 0 if we reach the end. */
if (++lg->next_hcall == LHCALL_RING_SIZE)
lg->next_hcall = 0;
- if (get_user(regs.eax, &lg->lguest_data->hcalls[n].eax)
- || get_user(regs.edx, &lg->lguest_data->hcalls[n].edx)
- || get_user(regs.ecx, &lg->lguest_data->hcalls[n].ecx)
- || get_user(regs.ebx, &lg->lguest_data->hcalls[n].ebx)) {
+ /* Copy the hypercall arguments into a local copy of
+ * the hcall_args struct. */
+ if (copy_from_user(&args, &lg->lguest_data->hcalls[n],
+ sizeof(struct hcall_args))) {
kill_guest(lg, "Fetching async hypercalls");
break;
}
- do_hcall(lg, ®s);
+ /* Do the hypercall, same as a normal one. */
+ do_hcall(lg, &args);
+
+ /* Mark the hypercall done. */
if (put_user(0xFF, &lg->lguest_data->hcall_status[n])) {
kill_guest(lg, "Writing result for async hypercall");
break;
}
- if (lg->dma_is_pending)
+ /* Stop doing hypercalls if they want to notify the Launcher:
+ * it needs to service this first. */
+ if (lg->pending_notify)
break;
}
}
+/* Last of all, we look at what happens first of all. The very first time the
+ * Guest makes a hypercall, we end up here to set things up: */
static void initialize(struct lguest *lg)
{
- u32 tsc_speed;
-
- if (lg->regs->eax != LHCALL_LGUEST_INIT) {
- kill_guest(lg, "hypercall %li before LGUEST_INIT",
- lg->regs->eax);
+ /* You can't do anything until you're initialized. The Guest knows the
+ * rules, so we're unforgiving here. */
+ if (lg->hcall->arg0 != LHCALL_LGUEST_INIT) {
+ kill_guest(lg, "hypercall %li before INIT", lg->hcall->arg0);
return;
}
- /* We only tell the guest to use the TSC if it's reliable. */
- if (boot_cpu_has(X86_FEATURE_CONSTANT_TSC) && !check_tsc_unstable())
- tsc_speed = tsc_khz;
- else
- tsc_speed = 0;
-
- lg->lguest_data = (struct lguest_data __user *)lg->regs->edx;
- /* We check here so we can simply copy_to_user/from_user */
- if (!lguest_address_ok(lg, lg->regs->edx, sizeof(*lg->lguest_data))) {
+ if (lguest_arch_init_hypercalls(lg))
kill_guest(lg, "bad guest page %p", lg->lguest_data);
- return;
- }
+
+ /* The Guest tells us where we're not to deliver interrupts by putting
+ * the range of addresses into "struct lguest_data". */
if (get_user(lg->noirq_start, &lg->lguest_data->noirq_start)
- || get_user(lg->noirq_end, &lg->lguest_data->noirq_end)
- /* We reserve the top pgd entry. */
- || put_user(4U*1024*1024, &lg->lguest_data->reserve_mem)
- || put_user(tsc_speed, &lg->lguest_data->tsc_khz)
- || put_user(lg->guestid, &lg->lguest_data->guestid))
+ || get_user(lg->noirq_end, &lg->lguest_data->noirq_end))
kill_guest(lg, "bad guest page %p", lg->lguest_data);
- /* This is the one case where the above accesses might have
- * been the first write to a Guest page. This may have caused
- * a copy-on-write fault, but the Guest might be referring to
- * the old (read-only) page. */
- guest_pagetable_clear_all(lg);
-}
+ /* We write the current time into the Guest's data page once so it can
+ * set its clock. */
+ write_timestamp(lg);
-/* Even if we go out to userspace and come back, we don't want to do
- * the hypercall again. */
-static void clear_hcall(struct lguest *lg)
-{
- lg->regs->trapnum = 255;
+ /* page_tables.c will also do some setup. */
+ page_table_guest_data_init(lg);
+
+ /* This is the one case where the above accesses might have been the
+ * first write to a Guest page. This may have caused a copy-on-write
+ * fault, but the old page might be (read-only) in the Guest
+ * pagetable. */
+ guest_pagetable_clear_all(lg);
}
+/*H:100
+ * Hypercalls
+ *
+ * Remember from the Guest, hypercalls come in two flavors: normal and
+ * asynchronous. This file handles both of types.
+ */
void do_hypercalls(struct lguest *lg)
{
+ /* Not initialized yet? This hypercall must do it. */
if (unlikely(!lg->lguest_data)) {
- if (lg->regs->trapnum == LGUEST_TRAP_ENTRY) {
- initialize(lg);
- clear_hcall(lg);
- }
+ /* Set up the "struct lguest_data" */
+ initialize(lg);
+ /* Hcall is done. */
+ lg->hcall = NULL;
return;
}
+ /* The Guest has initialized.
+ *
+ * Look in the hypercall ring for the async hypercalls: */
do_async_hcalls(lg);
- if (!lg->dma_is_pending && lg->regs->trapnum == LGUEST_TRAP_ENTRY) {
- do_hcall(lg, lg->regs);
- clear_hcall(lg);
+
+ /* If we stopped reading the hypercall ring because the Guest did a
+ * NOTIFY to the Launcher, we want to return now. Otherwise we do
+ * the hypercall. */
+ if (!lg->pending_notify) {
+ do_hcall(lg, lg->hcall);
+ /* Tricky point: we reset the hcall pointer to mark the
+ * hypercall as "done". We use the hcall pointer rather than
+ * the trap number to indicate a hypercall is pending.
+ * Normally it doesn't matter: the Guest will run again and
+ * update the trap number before we come back here.
+ *
+ * However, if we are signalled or the Guest sends I/O to the
+ * Launcher, the run_guest() loop will exit without running the
+ * Guest. When it comes back it would try to re-run the
+ * hypercall. */
+ lg->hcall = NULL;
}
}
+
+/* This routine supplies the Guest with time: it's used for wallclock time at
+ * initial boot and as a rough time source if the TSC isn't available. */
+void write_timestamp(struct lguest *lg)
+{
+ struct timespec now;
+ ktime_get_real_ts(&now);
+ if (copy_to_user(&lg->lguest_data->time, &now, sizeof(struct timespec)))
+ kill_guest(lg, "Writing timestamp");
+}