Re: [PATCH v10 02/25] x86/mm: define ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT

Hi Laurent,

Laurent Dufour [off-list ref] writes:

I'd suggest merging this patch with the one making changes to the
architectural fault handler towards the end of the series.

The Kconfig change is closely tied to the architectural support for SPF
and makes sense to be in a single patch.

If there's a good reason to keep them as separate patches, please move
the architecture Kconfig changes after the patch adding fault handler
changes.

It's better to enable the feature once the core infrastructure is merged
rather than at the beginning of the series to avoid potential bad
fallout from incomplete functionality during bisection.

All the comments here definitely hold for the arm64 patches that you
plan to include with the next update.

Thanks,
Punit

 
 #
 # Arch settings

On 08/05/2018 13:04, Punit Agrawal wrote:

Hi Laurent,

Laurent Dufour [off-list ref] writes:

quoted

Set ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT which turns on the
Speculative Page Fault handler when building for 64bit.

Cc: Thomas Gleixner <redacted>
Signed-off-by: Laurent Dufour <redacted>
---
 arch/x86/Kconfig | 1 +
 1 file changed, 1 insertion(+)

diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig
index d8983df5a2bc..ebdeb48e4a4a 100644
--- a/arch/x86/Kconfig
+++ b/arch/x86/Kconfig

@@ -30,6 +30,7 @@ config X86_64
 	select MODULES_USE_ELF_RELA
 	select X86_DEV_DMA_OPS
 	select ARCH_HAS_SYSCALL_WRAPPER
+	select ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT

I'd suggest merging this patch with the one making changes to the
architectural fault handler towards the end of the series.

The Kconfig change is closely tied to the architectural support for SPF
and makes sense to be in a single patch.

If there's a good reason to keep them as separate patches, please move
the architecture Kconfig changes after the patch adding fault handler
changes.

It's better to enable the feature once the core infrastructure is merged
rather than at the beginning of the series to avoid potential bad
fallout from incomplete functionality during bisection.

Indeed bisection was the reason why Andrew asked me to push the configuration
enablement on top of the series (https://lkml.org/lkml/2017/10/10/1229).

I also think it would be better to have the architecture enablement in on patch
but that would mean that the code will not be build when bisecting without the
latest patch adding the per architecture code.

I'm fine with the both options.

Andrew, what do you think would be the best here ?

Thanks,
Laurent.

All the comments here definitely hold for the arm64 patches that you
plan to include with the next update.

Thanks,
Punit

quoted

 
 #
 # Arch settings