* from frolicking through its parklike serenity. :*/
#include "lg.h"
-static int desc_ok(const struct desc_struct *gdt)
-{
- /* MBZ=0, P=1, DT=1 */
- return ((gdt->b & 0x00209000) == 0x00009000);
-}
-
-static int segment_present(const struct desc_struct *gdt)
-{
- return gdt->b & 0x8000;
-}
-
+/*H:600
+ * We've almost completed the Host; there's just one file to go!
+ *
+ * Segments & The Global Descriptor Table
+ *
+ * (That title sounds like a bad Nerdcore group. Not to suggest that there are
+ * any good Nerdcore groups, but in high school a friend of mine had a band
+ * called Joe Fish and the Chips, so there are definitely worse band names).
+ *
+ * To refresh: the GDT is a table of 8-byte values describing segments. Once
+ * set up, these segments can be loaded into one of the 6 "segment registers".
+ *
+ * GDT entries are passed around as "struct desc_struct"s, which like IDT
+ * entries are split into two 32-bit members, "a" and "b". One day, someone
+ * will clean that up, and be declared a Hero. (No pressure, I'm just saying).
+ *
+ * Anyway, the GDT entry contains a base (the start address of the segment), a
+ * limit (the size of the segment - 1), and some flags. Sounds simple, and it
+ * would be, except those zany Intel engineers decided that it was too boring
+ * to put the base at one end, the limit at the other, and the flags in
+ * between. They decided to shotgun the bits at random throughout the 8 bytes,
+ * like so:
+ *
+ * 0 16 40 48 52 56 63
+ * [ limit part 1 ][ base part 1 ][ flags ][li][fl][base ]
+ * mit ags part 2
+ * part 2
+ *
+ * As a result, this file contains a certain amount of magic numeracy. Let's
+ * begin.
+ */
+
+/* There are several entries we don't let the Guest set. The TSS entry is the
+ * "Task State Segment" which controls all kinds of delicate things. The
+ * LGUEST_CS and LGUEST_DS entries are reserved for the Switcher, and the
+ * the Guest can't be trusted to deal with double faults. */
static int ignored_gdt(unsigned int num)
{
return (num == GDT_ENTRY_TSS
|| num == GDT_ENTRY_DOUBLEFAULT_TSS);
}
-/* We don't allow removal of CS, DS or SS; it doesn't make sense. */
-static void check_segment_use(struct lguest *lg, unsigned int desc)
-{
- if (lg->regs->gs / 8 == desc)
- lg->regs->gs = 0;
- if (lg->regs->fs / 8 == desc)
- lg->regs->fs = 0;
- if (lg->regs->es / 8 == desc)
- lg->regs->es = 0;
- if (lg->regs->ds / 8 == desc
- || lg->regs->cs / 8 == desc
- || lg->regs->ss / 8 == desc)
- kill_guest(lg, "Removed live GDT entry %u", desc);
-}
-
+/*H:610 Once the GDT has been changed, we fix the new entries up a little. We
+ * don't care if they're invalid: the worst that can happen is a General
+ * Protection Fault in the Switcher when it restores a Guest segment register
+ * which tries to use that entry. Then we kill the Guest for causing such a
+ * mess: the message will be "unhandled trap 256". */
static void fixup_gdt_table(struct lguest *lg, unsigned start, unsigned end)
{
unsigned int i;
for (i = start; i < end; i++) {
- /* We never copy these ones to real gdt */
+ /* We never copy these ones to real GDT, so we don't care what
+ * they say */
if (ignored_gdt(i))
continue;
- /* We could fault in switch_to_guest if they are using
- * a removed segment. */
- if (!segment_present(&lg->gdt[i])) {
- check_segment_use(lg, i);
- continue;
- }
-
- if (!desc_ok(&lg->gdt[i]))
- kill_guest(lg, "Bad GDT descriptor %i", i);
-
- /* DPL 0 presumably means "for use by guest". */
+ /* Segment descriptors contain a privilege level: the Guest is
+ * sometimes careless and leaves this as 0, even though it's
+ * running at privilege level 1. If so, we fix it here. */
if ((lg->gdt[i].b & 0x00006000) == 0)
lg->gdt[i].b |= (GUEST_PL << 13);
- /* Set accessed bit, since gdt isn't writable. */
+ /* Each descriptor has an "accessed" bit. If we don't set it
+ * now, the CPU will try to set it when the Guest first loads
+ * that entry into a segment register. But the GDT isn't
+ * writable by the Guest, so bad things can happen. */
lg->gdt[i].b |= 0x00000100;
}
}
+/* This routine is called at boot or modprobe time for each CPU to set up the
+ * "constant" GDT entries for Guests running on that CPU. */
void setup_default_gdt_entries(struct lguest_ro_state *state)
{
struct desc_struct *gdt = state->guest_gdt;
unsigned long tss = (unsigned long)&state->guest_tss;
- /* Hypervisor segments. */
+ /* The hypervisor segments are full 0-4G segments, privilege level 0 */
gdt[GDT_ENTRY_LGUEST_CS] = FULL_EXEC_SEGMENT;
gdt[GDT_ENTRY_LGUEST_DS] = FULL_SEGMENT;
- /* This is the one which we *cannot* copy from guest, since tss
- is depended on this lguest_ro_state, ie. this cpu. */
+ /* The TSS segment refers to the TSS entry for this CPU, so we cannot
+ * copy it from the Guest. Forgive the magic flags */
gdt[GDT_ENTRY_TSS].a = 0x00000067 | (tss << 16);
gdt[GDT_ENTRY_TSS].b = 0x00008900 | (tss & 0xFF000000)
| ((tss >> 16) & 0x000000FF);
}
+/* This routine is called before the Guest is run for the first time. */
void setup_guest_gdt(struct lguest *lg)
{
+ /* Start with full 0-4G segments... */
lg->gdt[GDT_ENTRY_KERNEL_CS] = FULL_EXEC_SEGMENT;
lg->gdt[GDT_ENTRY_KERNEL_DS] = FULL_SEGMENT;
+ /* ...except the Guest is allowed to use them, so set the privilege
+ * level appropriately in the flags. */
lg->gdt[GDT_ENTRY_KERNEL_CS].b |= (GUEST_PL << 13);
lg->gdt[GDT_ENTRY_KERNEL_DS].b |= (GUEST_PL << 13);
}
-/* This is a fast version for the common case where only the three TLS entries
- * have changed. */
+/* Like the IDT, we never simply use the GDT the Guest gives us. We set up the
+ * GDTs for each CPU, then we copy across the entries each time we want to run
+ * a different Guest on that CPU. */
+
+/* A partial GDT load, for the three "thead-local storage" entries. Otherwise
+ * it's just like load_guest_gdt(). So much, in fact, it would probably be
+ * neater to have a single hypercall to cover both. */
void copy_gdt_tls(const struct lguest *lg, struct desc_struct *gdt)
{
unsigned int i;
gdt[i] = lg->gdt[i];
}
+/* This is the full version */
void copy_gdt(const struct lguest *lg, struct desc_struct *gdt)
{
unsigned int i;
+ /* The default entries from setup_default_gdt_entries() are not
+ * replaced. See ignored_gdt() above. */
for (i = 0; i < GDT_ENTRIES; i++)
if (!ignored_gdt(i))
gdt[i] = lg->gdt[i];
}
+/* This is where the Guest asks us to load a new GDT (LHCALL_LOAD_GDT). */
void load_guest_gdt(struct lguest *lg, unsigned long table, u32 num)
{
+ /* We assume the Guest has the same number of GDT entries as the
+ * Host, otherwise we'd have to dynamically allocate the Guest GDT. */
if (num > ARRAY_SIZE(lg->gdt))
kill_guest(lg, "too many gdt entries %i", num);
+ /* We read the whole thing in, then fix it up. */
lgread(lg, lg->gdt, table, num * sizeof(lg->gdt[0]));
fixup_gdt_table(lg, 0, ARRAY_SIZE(lg->gdt));
+ /* Mark that the GDT changed so the core knows it has to copy it again,
+ * even if the Guest is run on the same CPU. */
lg->changed |= CHANGED_GDT;
}
fixup_gdt_table(lg, GDT_ENTRY_TLS_MIN, GDT_ENTRY_TLS_MAX+1);
lg->changed |= CHANGED_GDT_TLS;
}
+
+/*
+ * With this, we have finished the Host.
+ *
+ * Five of the seven parts of our task are complete. You have made it through
+ * the Bit of Despair (I think that's somewhere in the page table code,
+ * myself).
+ *
+ * Next, we examine "make Switcher". It's short, but intense.
+ */
projects / ~andy / linux / blobdiff