1 /*P:600 The x86 architecture has segments, which involve a table of descriptors
2 * which can be used to do funky things with virtual address interpretation.
3 * We originally used to use segments so the Guest couldn't alter the
4 * Guest<->Host Switcher, and then we had to trim Guest segments, and restore
5 * for userspace per-thread segments, but trim again for on userspace->kernel
6 * transitions... This nightmarish creation was contained within this file,
7 * where we knew not to tread without heavy armament and a change of underwear.
9 * In these modern times, the segment handling code consists of simple sanity
10 * checks, and the worst you'll experience reading this code is butterfly-rash
11 * from frolicking through its parklike serenity. :*/
15 * We've almost completed the Host; there's just one file to go!
17 * Segments & The Global Descriptor Table
19 * (That title sounds like a bad Nerdcore group. Not to suggest that there are
20 * any good Nerdcore groups, but in high school a friend of mine had a band
21 * called Joe Fish and the Chips, so there are definitely worse band names).
23 * To refresh: the GDT is a table of 8-byte values describing segments. Once
24 * set up, these segments can be loaded into one of the 6 "segment registers".
26 * GDT entries are passed around as "struct desc_struct"s, which like IDT
27 * entries are split into two 32-bit members, "a" and "b". One day, someone
28 * will clean that up, and be declared a Hero. (No pressure, I'm just saying).
30 * Anyway, the GDT entry contains a base (the start address of the segment), a
31 * limit (the size of the segment - 1), and some flags. Sounds simple, and it
32 * would be, except those zany Intel engineers decided that it was too boring
33 * to put the base at one end, the limit at the other, and the flags in
34 * between. They decided to shotgun the bits at random throughout the 8 bytes,
38 * [ limit part 1 ][ base part 1 ][ flags ][li][fl][base ]
42 * As a result, this file contains a certain amount of magic numeracy. Let's
46 /* Is the descriptor the Guest wants us to put in OK?
48 * The flag which Intel says must be zero: must be zero. The descriptor must
49 * be present, (this is actually checked earlier but is here for thorougness),
50 * and the descriptor type must be 1 (a memory segment). */
51 static int desc_ok(const struct desc_struct *gdt)
53 return ((gdt->b & 0x00209000) == 0x00009000);
56 /* Is the segment present? (Otherwise it can't be used by the Guest). */
57 static int segment_present(const struct desc_struct *gdt)
59 return gdt->b & 0x8000;
62 /* There are several entries we don't let the Guest set. The TSS entry is the
63 * "Task State Segment" which controls all kinds of delicate things. The
64 * LGUEST_CS and LGUEST_DS entries are reserved for the Switcher, and the
65 * the Guest can't be trusted to deal with double faults. */
66 static int ignored_gdt(unsigned int num)
68 return (num == GDT_ENTRY_TSS
69 || num == GDT_ENTRY_LGUEST_CS
70 || num == GDT_ENTRY_LGUEST_DS
71 || num == GDT_ENTRY_DOUBLEFAULT_TSS);
74 /* If the Guest asks us to remove an entry from the GDT, we have to be careful.
75 * If one of the segment registers is pointing at that entry the Switcher will
76 * crash when it tries to reload the segment registers for the Guest.
78 * It doesn't make much sense for the Guest to try to remove its own code, data
79 * or stack segments while they're in use: assume that's a Guest bug. If it's
80 * one of the lesser segment registers using the removed entry, we simply set
81 * that register to 0 (unusable). */
82 static void check_segment_use(struct lguest *lg, unsigned int desc)
84 /* GDT entries are 8 bytes long, so we divide to get the index and
85 * ignore the bottom bits. */
86 if (lg->regs->gs / 8 == desc)
88 if (lg->regs->fs / 8 == desc)
90 if (lg->regs->es / 8 == desc)
92 if (lg->regs->ds / 8 == desc
93 || lg->regs->cs / 8 == desc
94 || lg->regs->ss / 8 == desc)
95 kill_guest(lg, "Removed live GDT entry %u", desc);
98 /*H:610 Once the GDT has been changed, we look through the changed entries and
99 * see if they're OK. If not, we'll call kill_guest() and the Guest will never
100 * get to use the invalid entries. */
101 static void fixup_gdt_table(struct lguest *lg, unsigned start, unsigned end)
105 for (i = start; i < end; i++) {
106 /* We never copy these ones to real GDT, so we don't care what
111 /* We could fault in switch_to_guest if they are using
112 * a removed segment. */
113 if (!segment_present(&lg->gdt[i])) {
114 check_segment_use(lg, i);
118 if (!desc_ok(&lg->gdt[i]))
119 kill_guest(lg, "Bad GDT descriptor %i", i);
121 /* Segment descriptors contain a privilege level: the Guest is
122 * sometimes careless and leaves this as 0, even though it's
123 * running at privilege level 1. If so, we fix it here. */
124 if ((lg->gdt[i].b & 0x00006000) == 0)
125 lg->gdt[i].b |= (GUEST_PL << 13);
127 /* Each descriptor has an "accessed" bit. If we don't set it
128 * now, the CPU will try to set it when the Guest first loads
129 * that entry into a segment register. But the GDT isn't
130 * writable by the Guest, so bad things can happen. */
131 lg->gdt[i].b |= 0x00000100;
135 /* This routine is called at boot or modprobe time for each CPU to set up the
136 * "constant" GDT entries for Guests running on that CPU. */
137 void setup_default_gdt_entries(struct lguest_ro_state *state)
139 struct desc_struct *gdt = state->guest_gdt;
140 unsigned long tss = (unsigned long)&state->guest_tss;
142 /* The hypervisor segments are full 0-4G segments, privilege level 0 */
143 gdt[GDT_ENTRY_LGUEST_CS] = FULL_EXEC_SEGMENT;
144 gdt[GDT_ENTRY_LGUEST_DS] = FULL_SEGMENT;
146 /* The TSS segment refers to the TSS entry for this CPU, so we cannot
147 * copy it from the Guest. Forgive the magic flags */
148 gdt[GDT_ENTRY_TSS].a = 0x00000067 | (tss << 16);
149 gdt[GDT_ENTRY_TSS].b = 0x00008900 | (tss & 0xFF000000)
150 | ((tss >> 16) & 0x000000FF);
153 /* This routine is called before the Guest is run for the first time. */
154 void setup_guest_gdt(struct lguest *lg)
156 /* Start with full 0-4G segments... */
157 lg->gdt[GDT_ENTRY_KERNEL_CS] = FULL_EXEC_SEGMENT;
158 lg->gdt[GDT_ENTRY_KERNEL_DS] = FULL_SEGMENT;
159 /* ...except the Guest is allowed to use them, so set the privilege
160 * level appropriately in the flags. */
161 lg->gdt[GDT_ENTRY_KERNEL_CS].b |= (GUEST_PL << 13);
162 lg->gdt[GDT_ENTRY_KERNEL_DS].b |= (GUEST_PL << 13);
165 /* Like the IDT, we never simply use the GDT the Guest gives us. We set up the
166 * GDTs for each CPU, then we copy across the entries each time we want to run
167 * a different Guest on that CPU. */
169 /* A partial GDT load, for the three "thead-local storage" entries. Otherwise
170 * it's just like load_guest_gdt(). So much, in fact, it would probably be
171 * neater to have a single hypercall to cover both. */
172 void copy_gdt_tls(const struct lguest *lg, struct desc_struct *gdt)
176 for (i = GDT_ENTRY_TLS_MIN; i <= GDT_ENTRY_TLS_MAX; i++)
180 /* This is the full version */
181 void copy_gdt(const struct lguest *lg, struct desc_struct *gdt)
185 /* The default entries from setup_default_gdt_entries() are not
186 * replaced. See ignored_gdt() above. */
187 for (i = 0; i < GDT_ENTRIES; i++)
192 /* This is where the Guest asks us to load a new GDT (LHCALL_LOAD_GDT). */
193 void load_guest_gdt(struct lguest *lg, unsigned long table, u32 num)
195 /* We assume the Guest has the same number of GDT entries as the
196 * Host, otherwise we'd have to dynamically allocate the Guest GDT. */
197 if (num > ARRAY_SIZE(lg->gdt))
198 kill_guest(lg, "too many gdt entries %i", num);
200 /* We read the whole thing in, then fix it up. */
201 lgread(lg, lg->gdt, table, num * sizeof(lg->gdt[0]));
202 fixup_gdt_table(lg, 0, ARRAY_SIZE(lg->gdt));
203 /* Mark that the GDT changed so the core knows it has to copy it again,
204 * even if the Guest is run on the same CPU. */
205 lg->changed |= CHANGED_GDT;
208 void guest_load_tls(struct lguest *lg, unsigned long gtls)
210 struct desc_struct *tls = &lg->gdt[GDT_ENTRY_TLS_MIN];
212 lgread(lg, tls, gtls, sizeof(*tls)*GDT_ENTRY_TLS_ENTRIES);
213 fixup_gdt_table(lg, GDT_ENTRY_TLS_MIN, GDT_ENTRY_TLS_MAX+1);
214 lg->changed |= CHANGED_GDT_TLS;
218 * With this, we have finished the Host.
220 * Five of the seven parts of our task are complete. You have made it through
221 * the Bit of Despair (I think that's somewhere in the page table code,
224 * Next, we examine "make Switcher". It's short, but intense.