Re: crash in entry.S restore_all, 2.6.12-rc2, x86, PAGEALLOC

[Linus Torvalds]

On Thu, 7 Apr 2005, Dave Jones wrote:
>
> On Thu, Apr 07, 2005 at 07:47:41AM -0700, Linus Torvalds wrote:
> 
>  > So the sysenter sequence might as well look like
>  > 
>  > 	pushl $(__USER_DS)	
>  > 	pushl %ebp
>  > 	sti
>  > 	pushfl
>  > 	..
>  > 
>  > which actually does three protected pushes thanks to the one-instruction 
>  > "interrupt shadow" after an sti.
> 
> Is this guaranteed on every x86 variant (or rather, every one
> that has SEP). ?

Well, since we only need two in this case, we don't care, but yes, it's 
supposed to be guaranteed by anything that calls itself an x86.

In fact, we _do_ depend on it in a few other sequences. Notably

	sti ; hlt

depends on the fact that an interrupt will always finish _after_ the hlt, 
and we'll never halt before the hlt (and then re-execute the hlt after the 
interrupt), and in

	sti ; iret

where we depend on the fact that we don't get recursive interrupt stacks 
(since we at that point have re-enabled the interrupt that happened).

Of course, if some future x86 decides that the interrupt shadow only
matters for special instructions (ie it's not so much a general interrupt
shadow as a "instruction combination"), I don't think Linux would care. I
really think there are only a very few valid sti-combinations, and I
suspect the above two are pretty much it.

(The other "magic" x86 behaviour is loading into the SS register, which
creates a one-cycle black hole after it. Linux shouldn't care, and in fact
nothing should care about it outside of old 16-bit non-protected-mode
programs, so I think that's another one that could be retired eventually)

		Linus
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/