On 17-06-19, 17:03, Julien Thierry wrote:
On 14/06/2019 04:07, Viresh Kumar wrote:
Hello,
Here is an attempt to backport arm64 spectre patches to v4.4 stable tree.
I have started this backport with Mark Rutland's backport of Spectre to 4.9 [1] and tried applying the upstream version of them over 4.4 and resolved conflicts by checking how they have been resolved in 4.9.
I had to pick few extra upstream patches to avoid unnecessary conflicts (upstream commit ids mentioned):
a842789837c0 arm64: remove duplicate macro __KERNEL__ check
I'm a bit unfamiliar with what gets or doesn't get backported. My understanding is that we try to backport only what's necessary to reduce the noise and potential introduction of issues in stable releases.
This commit is just a cleanup and (while valid) doesn't really seem necessary (and potential conflicts from its absence would easily be resolved IMO). So I'm just concerned that this doesn't constitute a candidate for back porting (someone can correct me if I'm wrong).
Dropped now.
64f8ebaf115b mm/kasan: add API to check memory regions bffe1baff5d5 arm64: kasan: instrument user memory access API 92406f0cc9e3 arm64: cpufeature: Add scope for capability check 9eb8a2cdf65c arm64: cputype info for Broadcom Vulcan 0d90718871fe arm64: cputype: Add MIDR values for Cavium ThunderX2 CPUs 98dd64f34f47 ARM: 8478/2: arm/arm64: add arm-smccc
I had to drop few patches as well as they weren't getting applied properly due to missing files/features (upstream commit id mentioned):
93f339ef4175 arm64: cpufeature: __this_cpu_has_cap() shouldn't stop early 3c31fa5a06b4 arm64: Run enable method for errata work arounds on late CPUs
Looking at this and at the patches that implement the BP callbacks, we need that patch or an equivalent, otherwise we won't be using the correct vectors for late CPUs...
I appreciate the code has changed, but it might be worth considering 6a6efbb45b7d95c84840010095367eb06a64f342 as a needed dependency for BP hardening.
Okay, I had to pick two more patches for a clean rebase.
d4a7e845dab5 arm64: Introduce cpu_die_early 7242dbf2e4da arm64: Move cpu_die_early to smp.c 545fe20330c3 arm64: Verify CPU errata work arounds on hotplugged CPU 0365babc6c1f arm64: Run enable method for errata work arounds on late CPUs
(You can fetch my tree again to get these commit ids)
6840bdd73d07 arm64: KVM: Use per-CPU vector when BP hardening is enabled
I don't believe we can do without this patch. Otherwise we're only using the vector that has no mitigation for kvm guests.
In v4.4, it looks like the contents of virt/kvm/arm/arm.c were contained in arch/arm/kvm/arm.c (yes, even for amr64). Are there other reasons this patch was not applying?
It was something other than this I believe, I have already used these paths for many other patches.
Anyway, KVM stuff is mostly dropped now, just that I had to keep the changes to arm-smccc.h from those patches.
I have updated the stable/v4.4.y/spectre branch with all the changes you suggested and pushed the earlier version to stable/v4.4.y/spectre-v1 branch.
Will it be possible for you to have a look at stable/v4.4.y/spectre branch to see if it is okay, so I can send the v2 version ? Don't want to spam list unnecessary with so many patches :)
Thanks for your help Julien, really appreciate it.