/*P:400 This contains run_guest() which actually calls into the Host<->Guest
* Switcher and analyzes the return, such as determining if the Guest wants the
- * Host to do something. This file also contains useful helper routines, and a
- * couple of non-obvious setup and teardown pieces which were implemented after
- * days of debugging pain. :*/
+ * Host to do something. This file also contains useful helper routines. :*/
#include <linux/module.h>
#include <linux/stringify.h>
#include <linux/stddef.h>
#include <linux/vmalloc.h>
#include <linux/cpu.h>
#include <linux/freezer.h>
+#include <linux/highmem.h>
#include <asm/paravirt.h>
-#include <asm/desc.h>
#include <asm/pgtable.h>
#include <asm/uaccess.h>
#include <asm/poll.h>
-#include <asm/highmem.h>
#include <asm/asm-offsets.h>
-#include <asm/i387.h>
#include "lg.h"
-/* Found in switcher.S */
-extern char start_switcher_text[], end_switcher_text[], switch_to_guest[];
-extern unsigned long default_idt_entries[];
-
-/* Every guest maps the core switcher code. */
-#define SHARED_SWITCHER_PAGES \
- DIV_ROUND_UP(end_switcher_text - start_switcher_text, PAGE_SIZE)
-/* Pages for switcher itself, then two pages per cpu */
-#define TOTAL_SWITCHER_PAGES (SHARED_SWITCHER_PAGES + 2 * NR_CPUS)
-
-/* We map at -4M for ease of mapping into the guest (one PTE page). */
-#define SWITCHER_ADDR 0xFFC00000
static struct vm_struct *switcher_vma;
static struct page **switcher_page;
-static int cpu_had_pge;
-static struct {
- unsigned long offset;
- unsigned short segment;
-} lguest_entry;
-
/* This One Big lock protects all inter-guest data structures. */
DEFINE_MUTEX(lguest_lock);
-static DEFINE_PER_CPU(struct lguest *, last_guest);
-
-/* FIXME: Make dynamic. */
-#define MAX_LGUEST_GUESTS 16
-struct lguest lguests[MAX_LGUEST_GUESTS];
-
-/* Offset from where switcher.S was compiled to where we've copied it */
-static unsigned long switcher_offset(void)
-{
- return SWITCHER_ADDR - (unsigned long)start_switcher_text;
-}
-
-/* This cpu's struct lguest_pages. */
-static struct lguest_pages *lguest_pages(unsigned int cpu)
-{
- return &(((struct lguest_pages *)
- (SWITCHER_ADDR + SHARED_SWITCHER_PAGES*PAGE_SIZE))[cpu]);
-}
+/*H:010 We need to set up the Switcher at a high virtual address. Remember the
+ * Switcher is a few hundred bytes of assembler code which actually changes the
+ * CPU to run the Guest, and then changes back to the Host when a trap or
+ * interrupt happens.
+ *
+ * The Switcher code must be at the same virtual address in the Guest as the
+ * Host since it will be running as the switchover occurs.
+ *
+ * Trying to map memory at a particular address is an unusual thing to do, so
+ * it's not a simple one-liner. */
static __init int map_switcher(void)
{
int i, err;
struct page **pagep;
+ /*
+ * Map the Switcher in to high memory.
+ *
+ * It turns out that if we choose the address 0xFFC00000 (4MB under the
+ * top virtual address), it makes setting up the page tables really
+ * easy.
+ */
+
+ /* We allocate an array of struct page pointers. map_vm_area() wants
+ * this, rather than just an array of pages. */
switcher_page = kmalloc(sizeof(switcher_page[0])*TOTAL_SWITCHER_PAGES,
GFP_KERNEL);
if (!switcher_page) {
goto out;
}
+ /* Now we actually allocate the pages. The Guest will see these pages,
+ * so we make sure they're zeroed. */
for (i = 0; i < TOTAL_SWITCHER_PAGES; i++) {
unsigned long addr = get_zeroed_page(GFP_KERNEL);
if (!addr) {
switcher_page[i] = virt_to_page(addr);
}
+ /* First we check that the Switcher won't overlap the fixmap area at
+ * the top of memory. It's currently nowhere near, but it could have
+ * very strange effects if it ever happened. */
+ if (SWITCHER_ADDR + (TOTAL_SWITCHER_PAGES+1)*PAGE_SIZE > FIXADDR_START){
+ err = -ENOMEM;
+ printk("lguest: mapping switcher would thwack fixmap\n");
+ goto free_pages;
+ }
+
+ /* Now we reserve the "virtual memory area" we want: 0xFFC00000
+ * (SWITCHER_ADDR). We might not get it in theory, but in practice
+ * it's worked so far. The end address needs +1 because __get_vm_area
+ * allocates an extra guard page, so we need space for that. */
switcher_vma = __get_vm_area(TOTAL_SWITCHER_PAGES * PAGE_SIZE,
- VM_ALLOC, SWITCHER_ADDR, VMALLOC_END);
+ VM_ALLOC, SWITCHER_ADDR, SWITCHER_ADDR
+ + (TOTAL_SWITCHER_PAGES+1) * PAGE_SIZE);
if (!switcher_vma) {
err = -ENOMEM;
printk("lguest: could not map switcher pages high\n");
goto free_pages;
}
+ /* This code actually sets up the pages we've allocated to appear at
+ * SWITCHER_ADDR. map_vm_area() takes the vma we allocated above, the
+ * kind of pages we're mapping (kernel pages), and a pointer to our
+ * array of struct pages. It increments that pointer, but we don't
+ * care. */
pagep = switcher_page;
err = map_vm_area(switcher_vma, PAGE_KERNEL, &pagep);
if (err) {
printk("lguest: map_vm_area failed: %i\n", err);
goto free_vma;
}
+
+ /* Now the Switcher is mapped at the right address, we can't fail!
+ * Copy in the compiled-in Switcher code (from <arch>_switcher.S). */
memcpy(switcher_vma->addr, start_switcher_text,
end_switcher_text - start_switcher_text);
- /* Fix up IDT entries to point into copied text. */
- for (i = 0; i < IDT_ENTRIES; i++)
- default_idt_entries[i] += switcher_offset();
-
- for_each_possible_cpu(i) {
- struct lguest_pages *pages = lguest_pages(i);
- struct lguest_ro_state *state = &pages->state;
-
- /* These fields are static: rest done in copy_in_guest_info */
- state->host_gdt_desc.size = GDT_SIZE-1;
- state->host_gdt_desc.address = (long)get_cpu_gdt_table(i);
- store_idt(&state->host_idt_desc);
- state->guest_idt_desc.size = sizeof(state->guest_idt)-1;
- state->guest_idt_desc.address = (long)&state->guest_idt;
- state->guest_gdt_desc.size = sizeof(state->guest_gdt)-1;
- state->guest_gdt_desc.address = (long)&state->guest_gdt;
- state->guest_tss.esp0 = (long)(&pages->regs + 1);
- state->guest_tss.ss0 = LGUEST_DS;
- /* No I/O for you! */
- state->guest_tss.io_bitmap_base = sizeof(state->guest_tss);
- setup_default_gdt_entries(state);
- setup_default_idt_entries(state, default_idt_entries);
-
- /* Setup LGUEST segments on all cpus */
- get_cpu_gdt_table(i)[GDT_ENTRY_LGUEST_CS] = FULL_EXEC_SEGMENT;
- get_cpu_gdt_table(i)[GDT_ENTRY_LGUEST_DS] = FULL_SEGMENT;
- }
-
- /* Initialize entry point into switcher. */
- lguest_entry.offset = (long)switch_to_guest + switcher_offset();
- lguest_entry.segment = LGUEST_CS;
-
printk(KERN_INFO "lguest: mapped switcher at %p\n",
switcher_vma->addr);
+ /* And we succeeded... */
return 0;
free_vma:
out:
return err;
}
+/*:*/
+/* Cleaning up the mapping when the module is unloaded is almost...
+ * too easy. */
static void unmap_switcher(void)
{
unsigned int i;
+ /* vunmap() undoes *both* map_vm_area() and __get_vm_area(). */
vunmap(switcher_vma->addr);
+ /* Now we just need to free the pages we copied the switcher into */
for (i = 0; i < TOTAL_SWITCHER_PAGES; i++)
__free_pages(switcher_page[i], 0);
+ kfree(switcher_page);
}
-/* IN/OUT insns: enough to get us past boot-time probing. */
-static int emulate_insn(struct lguest *lg)
-{
- u8 insn;
- unsigned int insnlen = 0, in = 0, shift = 0;
- unsigned long physaddr = guest_pa(lg, lg->regs->eip);
-
- /* This only works for addresses in linear mapping... */
- if (lg->regs->eip < lg->page_offset)
- return 0;
- lgread(lg, &insn, physaddr, 1);
-
- /* Operand size prefix means it's actually for ax. */
- if (insn == 0x66) {
- shift = 16;
- insnlen = 1;
- lgread(lg, &insn, physaddr + insnlen, 1);
- }
-
- switch (insn & 0xFE) {
- case 0xE4: /* in <next byte>,%al */
- insnlen += 2;
- in = 1;
- break;
- case 0xEC: /* in (%dx),%al */
- insnlen += 1;
- in = 1;
- break;
- case 0xE6: /* out %al,<next byte> */
- insnlen += 2;
- break;
- case 0xEE: /* out %al,(%dx) */
- insnlen += 1;
- break;
- default:
- return 0;
- }
-
- if (in) {
- /* Lower bit tells is whether it's a 16 or 32 bit access */
- if (insn & 0x1)
- lg->regs->eax = 0xFFFFFFFF;
- else
- lg->regs->eax |= (0xFFFF << shift);
- }
- lg->regs->eip += insnlen;
- return 1;
-}
-
+/*H:032
+ * Dealing With Guest Memory.
+ *
+ * Before we go too much further into the Host, we need to grok the routines
+ * we use to deal with Guest memory.
+ *
+ * When the Guest gives us (what it thinks is) a physical address, we can use
+ * the normal copy_from_user() & copy_to_user() on the corresponding place in
+ * the memory region allocated by the Launcher.
+ *
+ * But we can't trust the Guest: it might be trying to access the Launcher
+ * code. We have to check that the range is below the pfn_limit the Launcher
+ * gave us. We have to make sure that addr + len doesn't give us a false
+ * positive by overflowing, too. */
int lguest_address_ok(const struct lguest *lg,
unsigned long addr, unsigned long len)
{
return (addr+len) / PAGE_SIZE < lg->pfn_limit && (addr+len >= addr);
}
-/* Just like get_user, but don't let guest access lguest binary. */
-u32 lgread_u32(struct lguest *lg, unsigned long addr)
-{
- u32 val = 0;
-
- /* Don't let them access lguest binary */
- if (!lguest_address_ok(lg, addr, sizeof(val))
- || get_user(val, (u32 __user *)addr) != 0)
- kill_guest(lg, "bad read address %#lx", addr);
- return val;
-}
-
-void lgwrite_u32(struct lguest *lg, unsigned long addr, u32 val)
-{
- if (!lguest_address_ok(lg, addr, sizeof(val))
- || put_user(val, (u32 __user *)addr) != 0)
- kill_guest(lg, "bad write address %#lx", addr);
-}
-
-void lgread(struct lguest *lg, void *b, unsigned long addr, unsigned bytes)
+/* This routine copies memory from the Guest. Here we can see how useful the
+ * kill_lguest() routine we met in the Launcher can be: we return a random
+ * value (all zeroes) instead of needing to return an error. */
+void __lgread(struct lg_cpu *cpu, void *b, unsigned long addr, unsigned bytes)
{
- if (!lguest_address_ok(lg, addr, bytes)
- || copy_from_user(b, (void __user *)addr, bytes) != 0) {
+ if (!lguest_address_ok(cpu->lg, addr, bytes)
+ || copy_from_user(b, cpu->lg->mem_base + addr, bytes) != 0) {
/* copy_from_user should do this, but as we rely on it... */
memset(b, 0, bytes);
- kill_guest(lg, "bad read address %#lx len %u", addr, bytes);
- }
-}
-
-void lgwrite(struct lguest *lg, unsigned long addr, const void *b,
- unsigned bytes)
-{
- if (!lguest_address_ok(lg, addr, bytes)
- || copy_to_user((void __user *)addr, b, bytes) != 0)
- kill_guest(lg, "bad write address %#lx len %u", addr, bytes);
-}
-
-static void set_ts(void)
-{
- u32 cr0;
-
- cr0 = read_cr0();
- if (!(cr0 & 8))
- write_cr0(cr0|8);
-}
-
-static void copy_in_guest_info(struct lguest *lg, struct lguest_pages *pages)
-{
- if (__get_cpu_var(last_guest) != lg || lg->last_pages != pages) {
- __get_cpu_var(last_guest) = lg;
- lg->last_pages = pages;
- lg->changed = CHANGED_ALL;
+ kill_guest(cpu, "bad read address %#lx len %u", addr, bytes);
}
-
- /* These are pretty cheap, so we do them unconditionally. */
- pages->state.host_cr3 = __pa(current->mm->pgd);
- map_switcher_in_guest(lg, pages);
- pages->state.guest_tss.esp1 = lg->esp1;
- pages->state.guest_tss.ss1 = lg->ss1;
-
- /* Copy direct trap entries. */
- if (lg->changed & CHANGED_IDT)
- copy_traps(lg, pages->state.guest_idt, default_idt_entries);
-
- /* Copy all GDT entries but the TSS. */
- if (lg->changed & CHANGED_GDT)
- copy_gdt(lg, pages->state.guest_gdt);
- /* If only the TLS entries have changed, copy them. */
- else if (lg->changed & CHANGED_GDT_TLS)
- copy_gdt_tls(lg, pages->state.guest_gdt);
-
- lg->changed = 0;
}
-static void run_guest_once(struct lguest *lg, struct lguest_pages *pages)
+/* This is the write (copy into Guest) version. */
+void __lgwrite(struct lg_cpu *cpu, unsigned long addr, const void *b,
+ unsigned bytes)
{
- unsigned int clobber;
-
- copy_in_guest_info(lg, pages);
-
- /* Put eflags on stack, lcall does rest: suitable for iret return. */
- asm volatile("pushf; lcall *lguest_entry"
- : "=a"(clobber), "=b"(clobber)
- : "0"(pages), "1"(__pa(lg->pgdirs[lg->pgdidx].pgdir))
- : "memory", "%edx", "%ecx", "%edi", "%esi");
+ if (!lguest_address_ok(cpu->lg, addr, bytes)
+ || copy_to_user(cpu->lg->mem_base + addr, b, bytes) != 0)
+ kill_guest(cpu, "bad write address %#lx len %u", addr, bytes);
}
+/*:*/
-int run_guest(struct lguest *lg, unsigned long __user *user)
+/*H:030 Let's jump straight to the the main loop which runs the Guest.
+ * Remember, this is called by the Launcher reading /dev/lguest, and we keep
+ * going around and around until something interesting happens. */
+int run_guest(struct lg_cpu *cpu, unsigned long __user *user)
{
- while (!lg->dead) {
- unsigned int cr2 = 0; /* Damn gcc */
-
- /* Hypercalls first: we might have been out to userspace */
- do_hypercalls(lg);
- if (lg->dma_is_pending) {
- if (put_user(lg->pending_dma, user) ||
- put_user(lg->pending_key, user+1))
+ /* We stop running once the Guest is dead. */
+ while (!cpu->lg->dead) {
+ /* First we run any hypercalls the Guest wants done. */
+ if (cpu->hcall)
+ do_hypercalls(cpu);
+
+ /* It's possible the Guest did a NOTIFY hypercall to the
+ * Launcher, in which case we return from the read() now. */
+ if (cpu->pending_notify) {
+ if (put_user(cpu->pending_notify, user))
return -EFAULT;
- return sizeof(unsigned long)*2;
+ return sizeof(cpu->pending_notify);
}
+ /* Check for signals */
if (signal_pending(current))
return -ERESTARTSYS;
/* If Waker set break_out, return to Launcher. */
- if (lg->break_out)
+ if (cpu->break_out)
return -EAGAIN;
- maybe_do_interrupt(lg);
+ /* Check if there are any interrupts which can be delivered now:
+ * if so, this sets up the hander to be executed when we next
+ * run the Guest. */
+ maybe_do_interrupt(cpu);
+ /* All long-lived kernel loops need to check with this horrible
+ * thing called the freezer. If the Host is trying to suspend,
+ * it stops us. */
try_to_freeze();
- if (lg->dead)
+ /* Just make absolutely sure the Guest is still alive. One of
+ * those hypercalls could have been fatal, for example. */
+ if (cpu->lg->dead)
break;
- if (lg->halted) {
+ /* If the Guest asked to be stopped, we sleep. The Guest's
+ * clock timer or LHREQ_BREAK from the Waker will wake us. */
+ if (cpu->halted) {
set_current_state(TASK_INTERRUPTIBLE);
schedule();
continue;
}
+ /* OK, now we're ready to jump into the Guest. First we put up
+ * the "Do Not Disturb" sign: */
local_irq_disable();
- /* Even if *we* don't want FPU trap, guest might... */
- if (lg->ts)
- set_ts();
-
- /* Don't let Guest do SYSENTER: we can't handle it. */
- if (boot_cpu_has(X86_FEATURE_SEP))
- wrmsr(MSR_IA32_SYSENTER_CS, 0, 0);
-
- run_guest_once(lg, lguest_pages(raw_smp_processor_id()));
+ /* Actually run the Guest until something happens. */
+ lguest_arch_run_guest(cpu);
- /* Save cr2 now if we page-faulted. */
- if (lg->regs->trapnum == 14)
- cr2 = read_cr2();
- else if (lg->regs->trapnum == 7)
- math_state_restore();
-
- if (boot_cpu_has(X86_FEATURE_SEP))
- wrmsr(MSR_IA32_SYSENTER_CS, __KERNEL_CS, 0);
+ /* Now we're ready to be interrupted or moved to other CPUs */
local_irq_enable();
- switch (lg->regs->trapnum) {
- case 13: /* We've intercepted a GPF. */
- if (lg->regs->errcode == 0) {
- if (emulate_insn(lg))
- continue;
- }
- break;
- case 14: /* We've intercepted a page fault. */
- if (demand_page(lg, cr2, lg->regs->errcode))
- continue;
-
- /* If lguest_data is NULL, this won't hurt. */
- if (put_user(cr2, &lg->lguest_data->cr2))
- kill_guest(lg, "Writing cr2");
- break;
- case 7: /* We've intercepted a Device Not Available fault. */
- /* If they don't want to know, just absorb it. */
- if (!lg->ts)
- continue;
- break;
- case 32 ... 255: /* Real interrupt, fall thru */
- cond_resched();
- case LGUEST_TRAP_ENTRY: /* Handled at top of loop */
- continue;
- }
-
- if (deliver_trap(lg, lg->regs->trapnum))
- continue;
-
- kill_guest(lg, "unhandled trap %li at %#lx (%#lx)",
- lg->regs->trapnum, lg->regs->eip,
- lg->regs->trapnum == 14 ? cr2 : lg->regs->errcode);
+ /* Now we deal with whatever happened to the Guest. */
+ lguest_arch_handle_trap(cpu);
}
- return -ENOENT;
-}
-int find_free_guest(void)
-{
- unsigned int i;
- for (i = 0; i < MAX_LGUEST_GUESTS; i++)
- if (!lguests[i].tsk)
- return i;
- return -1;
-}
+ /* Special case: Guest is 'dead' but wants a reboot. */
+ if (cpu->lg->dead == ERR_PTR(-ERESTART))
+ return -ERESTART;
-static void adjust_pge(void *on)
-{
- if (on)
- write_cr4(read_cr4() | X86_CR4_PGE);
- else
- write_cr4(read_cr4() & ~X86_CR4_PGE);
+ /* The Guest is dead => "No such file or directory" */
+ return -ENOENT;
}
+/*H:000
+ * Welcome to the Host!
+ *
+ * By this point your brain has been tickled by the Guest code and numbed by
+ * the Launcher code; prepare for it to be stretched by the Host code. This is
+ * the heart. Let's begin at the initialization routine for the Host's lg
+ * module.
+ */
static int __init init(void)
{
int err;
+ /* Lguest can't run under Xen, VMI or itself. It does Tricky Stuff. */
if (paravirt_enabled()) {
- printk("lguest is afraid of %s\n", paravirt_ops.name);
+ printk("lguest is afraid of being a guest\n");
return -EPERM;
}
+ /* First we put the Switcher up in very high virtual memory. */
err = map_switcher();
if (err)
- return err;
+ goto out;
+ /* Now we set up the pagetable implementation for the Guests. */
err = init_pagetables(switcher_page, SHARED_SWITCHER_PAGES);
- if (err) {
- unmap_switcher();
- return err;
- }
- lguest_io_init();
+ if (err)
+ goto unmap;
+
+ /* We might need to reserve an interrupt vector. */
+ err = init_interrupts();
+ if (err)
+ goto free_pgtables;
+ /* /dev/lguest needs to be registered. */
err = lguest_device_init();
- if (err) {
- free_pagetables();
- unmap_switcher();
- return err;
- }
- lock_cpu_hotplug();
- if (cpu_has_pge) { /* We have a broader idea of "global". */
- cpu_had_pge = 1;
- on_each_cpu(adjust_pge, (void *)0, 0, 1);
- clear_bit(X86_FEATURE_PGE, boot_cpu_data.x86_capability);
- }
- unlock_cpu_hotplug();
+ if (err)
+ goto free_interrupts;
+
+ /* Finally we do some architecture-specific setup. */
+ lguest_arch_host_init();
+
+ /* All good! */
return 0;
+
+free_interrupts:
+ free_interrupts();
+free_pgtables:
+ free_pagetables();
+unmap:
+ unmap_switcher();
+out:
+ return err;
}
+/* Cleaning up is just the same code, backwards. With a little French. */
static void __exit fini(void)
{
lguest_device_remove();
+ free_interrupts();
free_pagetables();
unmap_switcher();
- lock_cpu_hotplug();
- if (cpu_had_pge) {
- set_bit(X86_FEATURE_PGE, boot_cpu_data.x86_capability);
- on_each_cpu(adjust_pge, (void *)1, 0, 1);
- }
- unlock_cpu_hotplug();
+
+ lguest_arch_host_fini();
}
+/*:*/
+/* The Host side of lguest can be a module. This is a nice way for people to
+ * play with it. */
module_init(init);
module_exit(fini);
MODULE_LICENSE("GPL");
git://ftp.safe.ca / safe / jmp / linux-2.6 / blobdiff