diff options
Diffstat (limited to 'drivers/lguest/x86/switcher_32.S')
-rw-r--r-- | drivers/lguest/x86/switcher_32.S | 71 |
1 files changed, 51 insertions, 20 deletions
diff --git a/drivers/lguest/x86/switcher_32.S b/drivers/lguest/x86/switcher_32.S index 1010b90b11fc..0af8baaa0d4a 100644 --- a/drivers/lguest/x86/switcher_32.S +++ b/drivers/lguest/x86/switcher_32.S | |||
@@ -6,6 +6,37 @@ | |||
6 | * are feeling invigorated and refreshed then the next, more challenging stage | 6 | * are feeling invigorated and refreshed then the next, more challenging stage |
7 | * can be found in "make Guest". :*/ | 7 | * can be found in "make Guest". :*/ |
8 | 8 | ||
9 | /*M:012 Lguest is meant to be simple: my rule of thumb is that 1% more LOC must | ||
10 | * gain at least 1% more performance. Since neither LOC nor performance can be | ||
11 | * measured beforehand, it generally means implementing a feature then deciding | ||
12 | * if it's worth it. And once it's implemented, who can say no? | ||
13 | * | ||
14 | * This is why I haven't implemented this idea myself. I want to, but I | ||
15 | * haven't. You could, though. | ||
16 | * | ||
17 | * The main place where lguest performance sucks is Guest page faulting. When | ||
18 | * a Guest userspace process hits an unmapped page we switch back to the Host, | ||
19 | * walk the page tables, find it's not mapped, switch back to the Guest page | ||
20 | * fault handler, which calls a hypercall to set the page table entry, then | ||
21 | * finally returns to userspace. That's two round-trips. | ||
22 | * | ||
23 | * If we had a small walker in the Switcher, we could quickly check the Guest | ||
24 | * page table and if the page isn't mapped, immediately reflect the fault back | ||
25 | * into the Guest. This means the Switcher would have to know the top of the | ||
26 | * Guest page table and the page fault handler address. | ||
27 | * | ||
28 | * For simplicity, the Guest should only handle the case where the privilege | ||
29 | * level of the fault is 3 and probably only not present or write faults. It | ||
30 | * should also detect recursive faults, and hand the original fault to the | ||
31 | * Host (which is actually really easy). | ||
32 | * | ||
33 | * Two questions remain. Would the performance gain outweigh the complexity? | ||
34 | * And who would write the verse documenting it? :*/ | ||
35 | |||
36 | /*M:011 Lguest64 handles NMI. This gave me NMI envy (until I looked at their | ||
37 | * code). It's worth doing though, since it would let us use oprofile in the | ||
38 | * Host when a Guest is running. :*/ | ||
39 | |||
9 | /*S:100 | 40 | /*S:100 |
10 | * Welcome to the Switcher itself! | 41 | * Welcome to the Switcher itself! |
11 | * | 42 | * |
@@ -88,7 +119,7 @@ ENTRY(switch_to_guest) | |||
88 | 119 | ||
89 | // All saved and there's now five steps before us: | 120 | // All saved and there's now five steps before us: |
90 | // Stack, GDT, IDT, TSS | 121 | // Stack, GDT, IDT, TSS |
91 | // And last of all the page tables are flipped. | 122 | // Then last of all the page tables are flipped. |
92 | 123 | ||
93 | // Yet beware that our stack pointer must be | 124 | // Yet beware that our stack pointer must be |
94 | // Always valid lest an NMI hits | 125 | // Always valid lest an NMI hits |
@@ -103,25 +134,25 @@ ENTRY(switch_to_guest) | |||
103 | lgdt LGUEST_PAGES_guest_gdt_desc(%eax) | 134 | lgdt LGUEST_PAGES_guest_gdt_desc(%eax) |
104 | 135 | ||
105 | // The Guest's IDT we did partially | 136 | // The Guest's IDT we did partially |
106 | // Move to the "struct lguest_pages" as well. | 137 | // Copy to "struct lguest_pages" as well. |
107 | lidt LGUEST_PAGES_guest_idt_desc(%eax) | 138 | lidt LGUEST_PAGES_guest_idt_desc(%eax) |
108 | 139 | ||
109 | // The TSS entry which controls traps | 140 | // The TSS entry which controls traps |
110 | // Must be loaded up with "ltr" now: | 141 | // Must be loaded up with "ltr" now: |
142 | // The GDT entry that TSS uses | ||
143 | // Changes type when we load it: damn Intel! | ||
111 | // For after we switch over our page tables | 144 | // For after we switch over our page tables |
112 | // It (as the rest) will be writable no more. | 145 | // That entry will be read-only: we'd crash. |
113 | // (The GDT entry TSS needs | ||
114 | // Changes type when we load it: damn Intel!) | ||
115 | movl $(GDT_ENTRY_TSS*8), %edx | 146 | movl $(GDT_ENTRY_TSS*8), %edx |
116 | ltr %dx | 147 | ltr %dx |
117 | 148 | ||
118 | // Look back now, before we take this last step! | 149 | // Look back now, before we take this last step! |
119 | // The Host's TSS entry was also marked used; | 150 | // The Host's TSS entry was also marked used; |
120 | // Let's clear it again, ere we return. | 151 | // Let's clear it again for our return. |
121 | // The GDT descriptor of the Host | 152 | // The GDT descriptor of the Host |
122 | // Points to the table after two "size" bytes | 153 | // Points to the table after two "size" bytes |
123 | movl (LGUEST_PAGES_host_gdt_desc+2)(%eax), %edx | 154 | movl (LGUEST_PAGES_host_gdt_desc+2)(%eax), %edx |
124 | // Clear the type field of "used" (byte 5, bit 2) | 155 | // Clear "used" from type field (byte 5, bit 2) |
125 | andb $0xFD, (GDT_ENTRY_TSS*8 + 5)(%edx) | 156 | andb $0xFD, (GDT_ENTRY_TSS*8 + 5)(%edx) |
126 | 157 | ||
127 | // Once our page table's switched, the Guest is live! | 158 | // Once our page table's switched, the Guest is live! |
@@ -131,7 +162,7 @@ ENTRY(switch_to_guest) | |||
131 | 162 | ||
132 | // The page table change did one tricky thing: | 163 | // The page table change did one tricky thing: |
133 | // The Guest's register page has been mapped | 164 | // The Guest's register page has been mapped |
134 | // Writable onto our %esp (stack) -- | 165 | // Writable under our %esp (stack) -- |
135 | // We can simply pop off all Guest regs. | 166 | // We can simply pop off all Guest regs. |
136 | popl %eax | 167 | popl %eax |
137 | popl %ebx | 168 | popl %ebx |
@@ -152,16 +183,15 @@ ENTRY(switch_to_guest) | |||
152 | addl $8, %esp | 183 | addl $8, %esp |
153 | 184 | ||
154 | // The last five stack slots hold return address | 185 | // The last five stack slots hold return address |
155 | // And everything needed to change privilege | 186 | // And everything needed to switch privilege |
156 | // Into the Guest privilege level of 1, | 187 | // From Switcher's level 0 to Guest's 1, |
157 | // And the stack where the Guest had last left it. | 188 | // And the stack where the Guest had last left it. |
158 | // Interrupts are turned back on: we are Guest. | 189 | // Interrupts are turned back on: we are Guest. |
159 | iret | 190 | iret |
160 | 191 | ||
161 | // There are two paths where we switch to the Host | 192 | // We treat two paths to switch back to the Host |
193 | // Yet both must save Guest state and restore Host | ||
162 | // So we put the routine in a macro. | 194 | // So we put the routine in a macro. |
163 | // We are on our way home, back to the Host | ||
164 | // Interrupted out of the Guest, we come here. | ||
165 | #define SWITCH_TO_HOST \ | 195 | #define SWITCH_TO_HOST \ |
166 | /* We save the Guest state: all registers first \ | 196 | /* We save the Guest state: all registers first \ |
167 | * Laid out just as "struct lguest_regs" defines */ \ | 197 | * Laid out just as "struct lguest_regs" defines */ \ |
@@ -194,7 +224,7 @@ ENTRY(switch_to_guest) | |||
194 | movl %esp, %eax; \ | 224 | movl %esp, %eax; \ |
195 | andl $(~(1 << PAGE_SHIFT - 1)), %eax; \ | 225 | andl $(~(1 << PAGE_SHIFT - 1)), %eax; \ |
196 | /* Save our trap number: the switch will obscure it \ | 226 | /* Save our trap number: the switch will obscure it \ |
197 | * (The Guest regs are not mapped here in the Host) \ | 227 | * (In the Host the Guest regs are not mapped here) \ |
198 | * %ebx holds it safe for deliver_to_host */ \ | 228 | * %ebx holds it safe for deliver_to_host */ \ |
199 | movl LGUEST_PAGES_regs_trapnum(%eax), %ebx; \ | 229 | movl LGUEST_PAGES_regs_trapnum(%eax), %ebx; \ |
200 | /* The Host GDT, IDT and stack! \ | 230 | /* The Host GDT, IDT and stack! \ |
@@ -210,9 +240,9 @@ ENTRY(switch_to_guest) | |||
210 | /* Switch to Host's GDT, IDT. */ \ | 240 | /* Switch to Host's GDT, IDT. */ \ |
211 | lgdt LGUEST_PAGES_host_gdt_desc(%eax); \ | 241 | lgdt LGUEST_PAGES_host_gdt_desc(%eax); \ |
212 | lidt LGUEST_PAGES_host_idt_desc(%eax); \ | 242 | lidt LGUEST_PAGES_host_idt_desc(%eax); \ |
213 | /* Restore the Host's stack where it's saved regs lie */ \ | 243 | /* Restore the Host's stack where its saved regs lie */ \ |
214 | movl LGUEST_PAGES_host_sp(%eax), %esp; \ | 244 | movl LGUEST_PAGES_host_sp(%eax), %esp; \ |
215 | /* Last the TSS: our Host is complete */ \ | 245 | /* Last the TSS: our Host is returned */ \ |
216 | movl $(GDT_ENTRY_TSS*8), %edx; \ | 246 | movl $(GDT_ENTRY_TSS*8), %edx; \ |
217 | ltr %dx; \ | 247 | ltr %dx; \ |
218 | /* Restore now the regs saved right at the first. */ \ | 248 | /* Restore now the regs saved right at the first. */ \ |
@@ -222,14 +252,15 @@ ENTRY(switch_to_guest) | |||
222 | popl %ds; \ | 252 | popl %ds; \ |
223 | popl %es | 253 | popl %es |
224 | 254 | ||
225 | // Here's where we come when the Guest has just trapped: | 255 | // The first path is trod when the Guest has trapped: |
226 | // (Which trap we'll see has been pushed on the stack). | 256 | // (Which trap it was has been pushed on the stack). |
227 | // We need only switch back, and the Host will decode | 257 | // We need only switch back, and the Host will decode |
228 | // Why we came home, and what needs to be done. | 258 | // Why we came home, and what needs to be done. |
229 | return_to_host: | 259 | return_to_host: |
230 | SWITCH_TO_HOST | 260 | SWITCH_TO_HOST |
231 | iret | 261 | iret |
232 | 262 | ||
263 | // We are lead to the second path like so: | ||
233 | // An interrupt, with some cause external | 264 | // An interrupt, with some cause external |
234 | // Has ajerked us rudely from the Guest's code | 265 | // Has ajerked us rudely from the Guest's code |
235 | // Again we must return home to the Host | 266 | // Again we must return home to the Host |
@@ -238,7 +269,7 @@ deliver_to_host: | |||
238 | // But now we must go home via that place | 269 | // But now we must go home via that place |
239 | // Where that interrupt was supposed to go | 270 | // Where that interrupt was supposed to go |
240 | // Had we not been ensconced, running the Guest. | 271 | // Had we not been ensconced, running the Guest. |
241 | // Here we see the cleverness of our stack: | 272 | // Here we see the trickness of run_guest_once(): |
242 | // The Host stack is formed like an interrupt | 273 | // The Host stack is formed like an interrupt |
243 | // With EIP, CS and EFLAGS layered. | 274 | // With EIP, CS and EFLAGS layered. |
244 | // Interrupt handlers end with "iret" | 275 | // Interrupt handlers end with "iret" |
@@ -263,7 +294,7 @@ deliver_to_host: | |||
263 | xorw %ax, %ax | 294 | xorw %ax, %ax |
264 | orl %eax, %edx | 295 | orl %eax, %edx |
265 | // Now the address of the handler's in %edx | 296 | // Now the address of the handler's in %edx |
266 | // We call it now: its "iret" takes us home. | 297 | // We call it now: its "iret" drops us home. |
267 | jmp *%edx | 298 | jmp *%edx |
268 | 299 | ||
269 | // Every interrupt can come to us here | 300 | // Every interrupt can come to us here |