Commit | Line | Data |
---|---|---|
f938d2c8 RR |
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. | |
8 | * | |
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. :*/ | |
d7e28ffe RR |
12 | #include "lg.h" |
13 | ||
bff672e6 RR |
14 | /*H:600 |
15 | * We've almost completed the Host; there's just one file to go! | |
16 | * | |
17 | * Segments & The Global Descriptor Table | |
18 | * | |
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). | |
22 | * | |
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". | |
25 | * | |
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). | |
29 | * | |
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, | |
35 | * like so: | |
36 | * | |
37 | * 0 16 40 48 52 56 63 | |
38 | * [ limit part 1 ][ base part 1 ][ flags ][li][fl][base ] | |
39 | * mit ags part 2 | |
40 | * part 2 | |
41 | * | |
42 | * As a result, this file contains a certain amount of magic numeracy. Let's | |
43 | * begin. | |
44 | */ | |
45 | ||
46 | /* Is the descriptor the Guest wants us to put in OK? | |
47 | * | |
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). */ | |
d7e28ffe RR |
51 | static int desc_ok(const struct desc_struct *gdt) |
52 | { | |
d7e28ffe RR |
53 | return ((gdt->b & 0x00209000) == 0x00009000); |
54 | } | |
55 | ||
bff672e6 | 56 | /* Is the segment present? (Otherwise it can't be used by the Guest). */ |
d7e28ffe RR |
57 | static int segment_present(const struct desc_struct *gdt) |
58 | { | |
59 | return gdt->b & 0x8000; | |
60 | } | |
61 | ||
bff672e6 RR |
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. */ | |
d7e28ffe RR |
66 | static int ignored_gdt(unsigned int num) |
67 | { | |
68 | return (num == GDT_ENTRY_TSS | |
69 | || num == GDT_ENTRY_LGUEST_CS | |
70 | || num == GDT_ENTRY_LGUEST_DS | |
71 | || num == GDT_ENTRY_DOUBLEFAULT_TSS); | |
72 | } | |
73 | ||
bff672e6 RR |
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. | |
77 | * | |
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). */ | |
d7e28ffe RR |
82 | static void check_segment_use(struct lguest *lg, unsigned int desc) |
83 | { | |
bff672e6 RR |
84 | /* GDT entries are 8 bytes long, so we divide to get the index and |
85 | * ignore the bottom bits. */ | |
d7e28ffe RR |
86 | if (lg->regs->gs / 8 == desc) |
87 | lg->regs->gs = 0; | |
88 | if (lg->regs->fs / 8 == desc) | |
89 | lg->regs->fs = 0; | |
90 | if (lg->regs->es / 8 == desc) | |
91 | lg->regs->es = 0; | |
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); | |
96 | } | |
f56a384e RR |
97 | /*:*/ |
98 | /*M:009 We wouldn't need to check for removal of in-use segments if we handled | |
99 | * faults in the Switcher. However, it's probably not a worthwhile | |
100 | * optimization. :*/ | |
d7e28ffe | 101 | |
bff672e6 RR |
102 | /*H:610 Once the GDT has been changed, we look through the changed entries and |
103 | * see if they're OK. If not, we'll call kill_guest() and the Guest will never | |
104 | * get to use the invalid entries. */ | |
d7e28ffe RR |
105 | static void fixup_gdt_table(struct lguest *lg, unsigned start, unsigned end) |
106 | { | |
107 | unsigned int i; | |
108 | ||
109 | for (i = start; i < end; i++) { | |
bff672e6 RR |
110 | /* We never copy these ones to real GDT, so we don't care what |
111 | * they say */ | |
d7e28ffe RR |
112 | if (ignored_gdt(i)) |
113 | continue; | |
114 | ||
115 | /* We could fault in switch_to_guest if they are using | |
116 | * a removed segment. */ | |
117 | if (!segment_present(&lg->gdt[i])) { | |
118 | check_segment_use(lg, i); | |
119 | continue; | |
120 | } | |
121 | ||
122 | if (!desc_ok(&lg->gdt[i])) | |
123 | kill_guest(lg, "Bad GDT descriptor %i", i); | |
124 | ||
bff672e6 RR |
125 | /* Segment descriptors contain a privilege level: the Guest is |
126 | * sometimes careless and leaves this as 0, even though it's | |
127 | * running at privilege level 1. If so, we fix it here. */ | |
d7e28ffe RR |
128 | if ((lg->gdt[i].b & 0x00006000) == 0) |
129 | lg->gdt[i].b |= (GUEST_PL << 13); | |
130 | ||
bff672e6 RR |
131 | /* Each descriptor has an "accessed" bit. If we don't set it |
132 | * now, the CPU will try to set it when the Guest first loads | |
133 | * that entry into a segment register. But the GDT isn't | |
134 | * writable by the Guest, so bad things can happen. */ | |
d7e28ffe RR |
135 | lg->gdt[i].b |= 0x00000100; |
136 | } | |
137 | } | |
138 | ||
bff672e6 RR |
139 | /* This routine is called at boot or modprobe time for each CPU to set up the |
140 | * "constant" GDT entries for Guests running on that CPU. */ | |
d7e28ffe RR |
141 | void setup_default_gdt_entries(struct lguest_ro_state *state) |
142 | { | |
143 | struct desc_struct *gdt = state->guest_gdt; | |
144 | unsigned long tss = (unsigned long)&state->guest_tss; | |
145 | ||
bff672e6 | 146 | /* The hypervisor segments are full 0-4G segments, privilege level 0 */ |
d7e28ffe RR |
147 | gdt[GDT_ENTRY_LGUEST_CS] = FULL_EXEC_SEGMENT; |
148 | gdt[GDT_ENTRY_LGUEST_DS] = FULL_SEGMENT; | |
149 | ||
bff672e6 RR |
150 | /* The TSS segment refers to the TSS entry for this CPU, so we cannot |
151 | * copy it from the Guest. Forgive the magic flags */ | |
d7e28ffe RR |
152 | gdt[GDT_ENTRY_TSS].a = 0x00000067 | (tss << 16); |
153 | gdt[GDT_ENTRY_TSS].b = 0x00008900 | (tss & 0xFF000000) | |
154 | | ((tss >> 16) & 0x000000FF); | |
155 | } | |
156 | ||
bff672e6 | 157 | /* This routine is called before the Guest is run for the first time. */ |
d7e28ffe RR |
158 | void setup_guest_gdt(struct lguest *lg) |
159 | { | |
bff672e6 | 160 | /* Start with full 0-4G segments... */ |
d7e28ffe RR |
161 | lg->gdt[GDT_ENTRY_KERNEL_CS] = FULL_EXEC_SEGMENT; |
162 | lg->gdt[GDT_ENTRY_KERNEL_DS] = FULL_SEGMENT; | |
bff672e6 RR |
163 | /* ...except the Guest is allowed to use them, so set the privilege |
164 | * level appropriately in the flags. */ | |
d7e28ffe RR |
165 | lg->gdt[GDT_ENTRY_KERNEL_CS].b |= (GUEST_PL << 13); |
166 | lg->gdt[GDT_ENTRY_KERNEL_DS].b |= (GUEST_PL << 13); | |
167 | } | |
168 | ||
bff672e6 RR |
169 | /* Like the IDT, we never simply use the GDT the Guest gives us. We set up the |
170 | * GDTs for each CPU, then we copy across the entries each time we want to run | |
171 | * a different Guest on that CPU. */ | |
172 | ||
173 | /* A partial GDT load, for the three "thead-local storage" entries. Otherwise | |
174 | * it's just like load_guest_gdt(). So much, in fact, it would probably be | |
175 | * neater to have a single hypercall to cover both. */ | |
d7e28ffe RR |
176 | void copy_gdt_tls(const struct lguest *lg, struct desc_struct *gdt) |
177 | { | |
178 | unsigned int i; | |
179 | ||
180 | for (i = GDT_ENTRY_TLS_MIN; i <= GDT_ENTRY_TLS_MAX; i++) | |
181 | gdt[i] = lg->gdt[i]; | |
182 | } | |
183 | ||
bff672e6 | 184 | /* This is the full version */ |
d7e28ffe RR |
185 | void copy_gdt(const struct lguest *lg, struct desc_struct *gdt) |
186 | { | |
187 | unsigned int i; | |
188 | ||
bff672e6 RR |
189 | /* The default entries from setup_default_gdt_entries() are not |
190 | * replaced. See ignored_gdt() above. */ | |
d7e28ffe RR |
191 | for (i = 0; i < GDT_ENTRIES; i++) |
192 | if (!ignored_gdt(i)) | |
193 | gdt[i] = lg->gdt[i]; | |
194 | } | |
195 | ||
bff672e6 | 196 | /* This is where the Guest asks us to load a new GDT (LHCALL_LOAD_GDT). */ |
d7e28ffe RR |
197 | void load_guest_gdt(struct lguest *lg, unsigned long table, u32 num) |
198 | { | |
bff672e6 RR |
199 | /* We assume the Guest has the same number of GDT entries as the |
200 | * Host, otherwise we'd have to dynamically allocate the Guest GDT. */ | |
d7e28ffe RR |
201 | if (num > ARRAY_SIZE(lg->gdt)) |
202 | kill_guest(lg, "too many gdt entries %i", num); | |
203 | ||
bff672e6 | 204 | /* We read the whole thing in, then fix it up. */ |
d7e28ffe RR |
205 | lgread(lg, lg->gdt, table, num * sizeof(lg->gdt[0])); |
206 | fixup_gdt_table(lg, 0, ARRAY_SIZE(lg->gdt)); | |
bff672e6 RR |
207 | /* Mark that the GDT changed so the core knows it has to copy it again, |
208 | * even if the Guest is run on the same CPU. */ | |
d7e28ffe RR |
209 | lg->changed |= CHANGED_GDT; |
210 | } | |
211 | ||
212 | void guest_load_tls(struct lguest *lg, unsigned long gtls) | |
213 | { | |
214 | struct desc_struct *tls = &lg->gdt[GDT_ENTRY_TLS_MIN]; | |
215 | ||
216 | lgread(lg, tls, gtls, sizeof(*tls)*GDT_ENTRY_TLS_ENTRIES); | |
217 | fixup_gdt_table(lg, GDT_ENTRY_TLS_MIN, GDT_ENTRY_TLS_MAX+1); | |
218 | lg->changed |= CHANGED_GDT_TLS; | |
219 | } | |
bff672e6 RR |
220 | |
221 | /* | |
222 | * With this, we have finished the Host. | |
223 | * | |
224 | * Five of the seven parts of our task are complete. You have made it through | |
225 | * the Bit of Despair (I think that's somewhere in the page table code, | |
226 | * myself). | |
227 | * | |
228 | * Next, we examine "make Switcher". It's short, but intense. | |
229 | */ |