Skip to contentSkip to conversation

i386: annotate the rest of entry.s::nmi

PATCH In review Chuck Ebbert 4 participants 7 messages Aug 10, 2006 – Aug 14, 2006

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 i386: annotate the rest of entry.s::nmi

Part of the NMI handler is missing annotations. Just moving the RING0_INT_FRAME macro fixes it. And additional comments should warn anyone changing this to recheck the annotations.

arch/i386/kernel/entry.S +8 -2
@@ -750,6 +750,7 @@ ENTRY(nmi)
 	cmpl $sysenter_entry,12(%esp)
 	je nmi_debug_stack_check
 nmi_stack_correct:
+	/* We have a RING0_INT_FRAME here */
 	pushl %eax
 	CFI_ADJUST_CFA_OFFSET 4
 	SAVE_ALL
@@ -760,9 +761,12 @@ nmi_stack_correct:
 	CFI_ENDPROC
 
 nmi_stack_fixup:
+	RING0_INT_FRAME
 	FIX_STACK(12,nmi_stack_correct, 1)
 	jmp nmi_stack_correct
+
 nmi_debug_stack_check:
+	/* We have a RING0_INT_FRAME here */
 	cmpw $__KERNEL_CS,16(%esp)
 	jne nmi_stack_correct
 	cmpl $debug,(%esp)
@@ -773,8 +777,10 @@ nmi_debug_stack_check:
 	jmp nmi_stack_correct
 
 nmi_16bit_stack:
-	RING0_INT_FRAME
-	/* create the pointer to lss back */
+	/* We have a RING0_INT_FRAME here.
+	 *
+	 * create the pointer to lss back
+	 */
 	pushl %ss
 	CFI_ADJUST_CFA_OFFSET 4
 	pushl %esp
Signature · 1 line
Chuck
#1 lore raw
Quoted 3 lines · Chuck Ebbert

Part of the NMI handler is missing annotations. Just moving the RING0_INT_FRAME macro fixes it. And additional comments should warn anyone changing this to recheck the annotations.

Added thanks.
-Andi

#2 lore raw
In reply to Chuck Ebbert
Quoted 4 lines
Quoted 1 line
Quoted 1 line

Chuck Ebbert <[email protected]> 10.08.06 06:59 >>>

Part of the NMI handler is missing annotations. Just moving the RING0_INT_FRAME macro fixes it. And additional comments should warn anyone changing this to recheck the annotations.

I have to admit that I can't see the value of this movement; the code sequence in question was left un-annotated intentionally. The point is that the push-es in FIX_STACK() aren't annotated, so things won't be correct at those points anyway. Furthermore, getting interrupted in any way while in this code path is going to kill the system (and is just impossible AFAICT), so annotations there also seem worthless. (I am actually surprised that I annotated the code after nmi_16bit_stack, as that's similarly constrained; even more, as I'm now looking at it, this code seems outright wrong in using iret since that unmasks NMIs - Stas, is your pending adjustment to the 16-bit stack handling going to overcome this?)

The added comments certainly might be helpful, though there are more places where frame state gets "inherited" across labels.

Jan

#3 lore raw
Quoted 6 lines · Jan Beulich
Quoted 4 lines
Quoted 1 line
Quoted 1 line

Chuck Ebbert <[email protected]> 10.08.06 06:59 >>>

Part of the NMI handler is missing annotations. Just moving the RING0_INT_FRAME macro fixes it. And additional comments should warn anyone changing this to recheck the annotations.

I have to admit that I can't see the value of this movement;

Should I drop it again?

-Andi

#4 lore raw
Quoted 10 lines
Quoted 1 line
Quoted 1 line

Andi Kleen <[email protected]> 10.08.06 10:26 >>>

Quoted 6 lines · Jan Beulich
Quoted 4 lines
Quoted 1 line
Quoted 1 line

Chuck Ebbert <[email protected]> 10.08.06 06:59 >>>

Part of the NMI handler is missing annotations. Just moving the RING0_INT_FRAME macro fixes it. And additional comments should warn anyone changing this to recheck the annotations.

I have to admit that I can't see the value of this movement;

Should I drop it again?

I would suggest so, but would also want to wait for Chuck to perhaps provide some background on why he felt these annotations were useful.

Jan

#5 lore raw
In reply to Jan Beulich

Hello.

Quoted 4 lines · Jan Beulich

even more, as I'm now looking at it, this code seems outright wrong in using iret since that unmasks NMIs - Stas, is your pending adjustment to the 16-bit stack handling going to overcome this?)

No, it leaves the NMI path almost untouched. But what exactly problem do you see with this iret? If it unmasks NMI, then it does so for reason, which is a return from an NMI handler. What exactly is wrong with it?

#6 lore raw
Quoted 9 lines
Quoted 4 lines

even more, as I'm now looking at it, this code seems outright wrong in using iret since that unmasks NMIs - Stas, is your pending adjustment to the 16-bit stack handling going to overcome this?)

No, it leaves the NMI path almost untouched. But what exactly problem do you see with this iret? If it unmasks NMI, then it does so for reason, which is a return from an NMI handler. What exactly is wrong with it?

Oh, yes, you're right, I didn't pay attention to the second do_nmi call in that path.

Sorry, Jan