[v2,0/5] arm64: IRQ priority masking and Pseudo-NMI fixes

Message

Julien Thierry April 29, 2019, 4 p.m. UTC

Hi,

[Changing the title to make it reflex more the status of the series.]

Version one[1] of this series attempted to fix the issue reported by
Zenghui[2] when using the function_graph tracer with IRQ priority
masking.

Since then, I realized that priority masking and the use of Pseudo-NMIs
was more broken than I thought.

* Patch 1 is just some rework
* Patch 2 replaces the previous irqflags tracing "simplification" with
  a proper fix when tracing Pseudo-NMI
* Patch 3 fixes the function_graph issue when using priority masking
* Patch 4 introduces some debug to hopefully avoid breaking things in
  the future
* Patch 5 is a rebased version of the patch sent by Wei Li[3] fixing
  an error that can happen during on some platform using the priority
  masking feature

Changes since V1 [1]:
- Fix possible race condition between NMI and trace irqflags
- Simplify the representation of PSR.I in the PMR value
- Include Wei Li's fix
- Rebase on v5.1-rc7

[1] https://marc.info/?l=linux-arm-kernel&m=155542458004480&w=2
[2] https://www.spinics.net/lists/arm-kernel/msg716956.html
[3] https://www.spinics.net/lists/arm-kernel/msg722171.html

Cheers,

Julien

-->

Julien Thierry (4):
  arm64: Do not enable IRQs for ct_user_exit
  arm64: Fix interrupt tracing in the presence of NMIs
  arm64: Fix incorrect irqflag restore for priority masking
  arm64: irqflags: Introduce explicit debugging for IRQ priorities

Wei Li (1):
  arm64: fix kernel stack overflow in kdump capture kernel

 arch/arm64/Kconfig                  | 11 +++++
 arch/arm64/include/asm/arch_gicv3.h |  4 +-
 arch/arm64/include/asm/assembler.h  |  9 ++++
 arch/arm64/include/asm/daifflags.h  | 31 +++++++++++---
 arch/arm64/include/asm/irqflags.h   | 83 +++++++++++++++++++------------------
 arch/arm64/include/asm/kvm_host.h   |  4 +-
 arch/arm64/include/asm/ptrace.h     | 10 ++++-
 arch/arm64/kernel/entry.S           | 76 +++++++++++++++++++++++++--------
 arch/arm64/kernel/irq.c             | 26 ++++++++++++
 arch/arm64/kernel/process.c         |  2 +-
 arch/arm64/kernel/smp.c             |  6 +--
 arch/arm64/kvm/hyp/switch.c         |  2 +-
 drivers/irqchip/irq-gic-v3.c        |  6 +++
 kernel/irq/irqdesc.c                |  8 +++-
 14 files changed, 202 insertions(+), 76 deletions(-)

--
1.9.1

Comments

Will Deacon May 23, 2019, 4:51 p.m. UTC | #1

Hi Julien,

On Mon, Apr 29, 2019 at 05:00:02PM +0100, Julien Thierry wrote:
> [Changing the title to make it reflex more the status of the series.]
> 
> Version one[1] of this series attempted to fix the issue reported by
> Zenghui[2] when using the function_graph tracer with IRQ priority
> masking.
> 
> Since then, I realized that priority masking and the use of Pseudo-NMIs
> was more broken than I thought.

Do you plan to respin this in light of Marc's comments?

Will

Will Deacon June 5, 2019, 4:07 p.m. UTC | #2

Hi again, Julien,

On Thu, May 23, 2019 at 05:51:55PM +0100, Will Deacon wrote:
> On Mon, Apr 29, 2019 at 05:00:02PM +0100, Julien Thierry wrote:
> > [Changing the title to make it reflex more the status of the series.]
> > 
> > Version one[1] of this series attempted to fix the issue reported by
> > Zenghui[2] when using the function_graph tracer with IRQ priority
> > masking.
> > 
> > Since then, I realized that priority masking and the use of Pseudo-NMIs
> > was more broken than I thought.
> 
> Do you plan to respin this in light of Marc's comments?

For now, I marked this as depending on BROKEN in mainline, but please can
you look at respinning these fixes so that we can get things fixed properly
for 5.3?

Thanks,

Will