Re: [PATCH V5 01/50] x86/entry: Add fence for kernel entry swapgs in paranoid_entry()

[Lai Jiangshan]

On 2021/11/18 23:54, Peter Zijlstra wrote:

> I'm confused, shouldn't the LFENCE be between SWAPGS and future uses of
> GS prefix?

I'm wrong a again.

I once thought "it should be followed with serializing operations such
as SWITCH_TO_KERNEL_CR3", and tglx corrected me:

https://lore.kernel.org/lkml/875yumbgox.ffs@tglx/

> It does not matter whether the *serializing* is before or after

And in my brain, it was incorrectly stored as:
   It does not matter whether the *fence* is before or after.

I will update patch1 and the corresponding C code in later patches.

Patch 1 in V4 is correct, but not as good as Borislav Petkov pointed out
that it has duplicated FENCE_SWAPGS_KERNEL_ENTRY.

I will change it as

	rdmsr
	if (need_swapgs) {
		swapgs
		set ebx/return value
	}
	FENCE_SWAPGS_KERNEL_ENTRY

> In the old code, before 96b2371413e8f, we had:
>
> 	swapgs
> 	SAVE_AND_SWITCH_TO_KERNEL_CR3
> 	FENCE_SWAPGS_KERNEL_ENTRY
>
> 	// %gs user comes here..
>
> And the comment made sense, since if SAVE_AND_SWITCH_TO_KERNEL_CR3 would
> imply an unconditional CR3 write, the LFENCE would not be needed.
>
> Then along gomes 96b2371413e8f and changes the order to:
>
> 	SAVE_AND_SWITCH_TO_KERNEL_CR3
> 	swapgs
> 	FENCE_SWAPGS_KERNEL_ENTRY
> 	// %gs user comes here..
>
> But now the comment is crazy talk, because even if the CR3 write were
> unconditional, it'd be pointless, since it's not after SWAPGS, but we
> still have the LFENCE in the right place.

I think the comments also make sense.

If CR3 write were unconditional before swapgs, no fence is needed after
swapgs since cr3 write is serializing.

> But now you want to make it:
>
> 	SAVE_AND_SWITCH_TO_KERNEL_CR3
> 	FENCE_SWAPGS_KERNEL_ENTRY
> 	swapgs
> 	// %gs user comes here..
>
> And there's nothing left and speculation can use the old %gs for our
> user and things go sideways. Hmm?
>
>
> (on a completely unrelated note, I find KERNEL_ENTRY and USER_ENTRY
> utterly confusing)