Re: [PATCH 0/4] Remove interrupt stack table usage from x86_64 kernel (v2)

[Ingo Molnar <mingo@elte.hu>]

* Avi Kivity <avi@redhat.com> wrote:

> The interrupt stack table (IST) mechanism is the only thing preventing
> kvm from deferring saving and reloading of some significant state.  It
> is also somewhat complicated.
> 
> Remove it by switching the special exceptions to use the normal irqstack.
> 
> Changes from v1:
> - rebase on tip/master
> - as a step, consolidate stack switching into a single macro
> 
> Jeremy, Xen is also affected; please review.
> 
> Avi Kivity (4):
>   x86: drop the use of the tss interrupt stack table (IST)
>   x86: Consolidate irq stack switching to a single macro
>   x86: Make interrupt stack switching atomic
>   x86: Move NMI back to interrupt stack
> 
>  arch/x86/include/asm/desc.h      |   12 -----
>  arch/x86/include/asm/page_64.h   |    7 ---
>  arch/x86/include/asm/pda.h       |    2 +-
>  arch/x86/include/asm/processor.h |   11 ----
>  arch/x86/kernel/asm-offsets_64.c |    1 -
>  arch/x86/kernel/cpu/common.c     |   35 --------------
>  arch/x86/kernel/dumpstack_64.c   |   96 --------------------------------------
>  arch/x86/kernel/entry_64.S       |   89 ++++++++++-------------------------
>  arch/x86/kernel/traps.c          |   12 ++--
>  9 files changed, 33 insertions(+), 232 deletions(-)

applied to tip/x86/irq, thanks Avi!

They have the following commit IDs, and they are also in tip/master:

 921e521: x86: move NMI back to interrupt stack
 36ef6c9: x86: make interrupt stack switching atomic
 dd64891: x86: consolidate irq stack switching to a single macro
 955a368: x86: drop the use of the tss interrupt stack table (IST)

I also started testing them in tip-qa.

I added the standard Impact-lines that we do in the x86 tree. Note that 
this patch:

 dd64891: x86: consolidate irq stack switching to a single macro

isnt just consolidating IRQ entry assembly code, it is also changing the 
paranoidentry macros to do IRQ stack entries - and hence switches all but 
the NMI critical exception entries sequences over to the IRQ stack. Your 
later patch:

 921e521: x86: move NMI back to interrupt stack

covers the NMI entry code too.

Please double-check that we indeed now have all the critical exceptions on 
the IRQ stack (they are all rare so testing alone wont show this), and 
please also double-check that we dont have more exceptions and entry 
callpaths on the IRQ stack than what we wanted. For example on a 
preemptible kernel (or in any codepath that calls schedule()) it is fatal 
to be on the IRQ stack, so this has to be very accurately coded.

	Ingo