Skip to contentSkip to conversation

i386 Vmi code patching

RFC PATCH Chuck Ebbert 2 participants 2 messages Mar 28, 2006

lore
Share & tools
lore
#0 lore raw

In-Reply-To: <[email protected]>

Quoted 10 lines · Andi Kleen
Quoted 8 lines · Zachary Amsden

The VMI ROM detection and code patching mechanism is illustrated in setup.c. There ROM is a binary block published by the hypervisor, and and there are certainly implications of this. ROMs certainly have a history of being proprietary, very differently licensed pieces of software, and mostly under non-free licenses. Before jumping to the conclusion that this is a bad thing, let us consider more carefully why hiding the interface layer to the hypervisor is actually a good thing.

Quoted 5 lines

How about you fix all these issues you describe here first and then submit it again?

The disassembly stuff indeed doesn't look like something that belongs in the kernel.

I think they put the disassembler in there as a joke. ;)

It's not necessary for fixing up the call site, anyway. Something like this should work, assuming there is always a call in every vmi translation.

/* Now, measure and emit the vmi translation sequence */
#define vmi_translation_start \
.pushsection .vmi.translation,"ax"; \
781:;
#define vmi_translation_finish \
- 782:; \
+ 783:; \
.popsection;
#define vmi_translation_begin 781b
-#define vmi_translation_end 782b
+#define vmi_call_location 782b
+#define vmi_translation_end 783b
#define vmi_translation_len (vmi_translation_end - vmi_translation_begin)
+#define vmi_call_offset (vmi_call_location - vmi_translation_begin)

#define vmi_call(name) \
- call .+5+name
+ 782: call .+5+name

#define vmi_annotate(name) \
.pushsection .vmi.annotation,"a"; \
.align 4; \
.long name; \
.long vmi_padded_begin; \
.long vmi_translation_begin; \
.byte vmi_padded_len; \
.byte vmi_translation_len; \
.byte vmi_pad_total; \
- .byte 0; \
+ .byte vmi_call_offset; \
.popsection;

struct vmi_annotation {
unsigned long vmi_call;
unsigned char *nativeEIP;
unsigned char *translationEIP;
unsigned char native_size;
unsigned char translation_size;
char nop_size;
- unsigned char pad;
+ unsigned char call_offset;
};

static void fixup_translation(struct vmi_annotation *a) {
unsigned char *c, *start, *end;
int left;

memcpy(a->nativeEIP, a->translationEIP, a->translation_size);
+ patch_call_site(a, a->nativeEIP + a->call_offset);
- start = a->nativeEIP;
- end = a->nativeEIP + a->translation_size;
-
- for (c = start; c < end;) {
- switch(*c) {
- case MNEM_CALL_NEAR:
- patch_call_site(a, c);
- c+=5;
- break;
-
- case MNEM_PUSH_I:
- c+=5;
- break;
-
- case MNEM_PUSH_IB:
- c+=2;
- break;
-
- case MNEM_PUSH_EAX:
- case MNEM_PUSH_ECX:
- case MNEM_PUSH_EDX:
- case MNEM_PUSH_EBX:
- case MNEM_PUSH_EBP:
- case MNEM_PUSH_ESI:
- case MNEM_PUSH_EDI:
- c+=1;
- break;
-
- case MNEM_LEA:
- BUG_ON(*(c+1) != 0x64); /* [--][--]+disp8, %esp */
- BUG_ON(*(c+2) != 0x24); /* none + %esp */
- c+=4;
- break;
-
- default:
- /*
- * Don't printk - it may acquire spinlocks with
- * partially completed VMI translations, causing
- * nuclear meltdown of the core.
- */
- BUG();
- return;
- }
- }

Signature · 2 lines
Chuck
"Penguins don't come from next door, they come from the Antarctic!"
#1 lore raw
Quoted 22 lines · Chuck Ebbert

In-Reply-To: <[email protected]>

Quoted 11 lines · Andi Kleen
Quoted 9 lines · Zachary Amsden

The VMI ROM detection and code patching mechanism is illustrated in setup.c. There ROM is a binary block published by the hypervisor, and and there are certainly implications of this. ROMs certainly have a history of being proprietary, very differently licensed pieces of software, and mostly under non-free licenses. Before jumping to the conclusion that this is a bad thing, let us consider more carefully why hiding the interface layer to the hypervisor is actually a good thing.

Quoted 5 lines

How about you fix all these issues you describe here first and then submit it again?

The disassembly stuff indeed doesn't look like something that belongs in the kernel.

Quoted 6 lines

I think they put the disassembler in there as a joke. ;)

It's not necessary for fixing up the call site, anyway. Something like this should work, assuming there is always a call in every vmi translation.

Very good observation. The code you illustrate is roughly what the patch mechanism used to do. The disassembly serves one purpose, which is incomplete in our current implementation. It provides live register information and constant propagation that can be used by an inliner in the VMI layer to inline hypervisor calls. This information gets encoded naturally in the push sequence before the call instruction.

But considering it is currently incomplete, and most of the benefit can be had be special casing just 4 VMI calls to allow selective inlining, it does seem like a lot of complexity for little payoff. I really don't like special casing, but in this scenario, it does seem to make sense. And of the 4 VMI calls, only one (SetInterruptMask) takes an input, and it takes that input in a fixed register.

Based on suggestions from Anthony Liguori and others, we are going to drop this extra complexity - we can get to hypervisor inlining later. Right now having the simplest and cleanest interface is more important. Actually, adding the required padding for inlining is quite easy:

#define FILL(n) ".fill " #n "-1,1,0x66; nop"

static inline void local_irq_restore(const unsigned long flags)

{
       vmi_wrap_call(
               SetInterruptMask, "pushl %0; popfl"  FILL(6),
               VMI_NO_OUTPUT,
               1, VMI_IREG1 (flags),
               XCONC("cc", "memory"));
}

Now you have 8 bytes to overwrite, which is sufficient for byte arithmetic operations to a memory address, plus a segment override. This seems like a much simpler solution than run-time disassembly.

Review

Quoted 10 lines
(quoted diff) +0 -9
-				/*
-				 * Don't printk - it may acquire spinlocks with
-				 * partially completed VMI translations, causing
-				 * nuclear meltdown of the core.
- 				 */
-				BUG();
-				return;
-		}
-	}

This part was a joke ;)