aboutsummaryrefslogtreecommitdiffstats
path: root/tools/testing/selftests
diff options
context:
space:
mode:
authorMarc Zyngier <marc.zyngier@arm.com>2019-07-05 23:35:56 +0100
committerMarc Zyngier <marc.zyngier@arm.com>2019-07-08 16:29:48 +0100
commit1e0cf16cdad1ba53e9eeee8746fe57de42f20c97 (patch)
tree664d34bdb34d0e9e7ee193932c225ed49a4db80e /tools/testing/selftests
parentKVM: arm64: Migrate _elx sysreg accessors to msr_s/mrs_s (diff)
downloadlinux-dev-1e0cf16cdad1ba53e9eeee8746fe57de42f20c97.tar.xz
linux-dev-1e0cf16cdad1ba53e9eeee8746fe57de42f20c97.zip
KVM: arm/arm64: Initialise host's MPIDRs by reading the actual register
As part of setting up the host context, we populate its MPIDR by using cpu_logical_map(). It turns out that contrary to arm64, cpu_logical_map() on 32bit ARM doesn't return the *full* MPIDR, but a truncated version. This leaves the host MPIDR slightly corrupted after the first run of a VM, since we won't correctly restore the MPIDR on exit. Oops. Since we cannot trust cpu_logical_map(), let's adopt a different strategy. We move the initialization of the host CPU context as part of the per-CPU initialization (which, in retrospect, makes a lot of sense), and directly read the MPIDR from the HW. This is guaranteed to work on both arm and arm64. Reported-by: Andre Przywara <Andre.Przywara@arm.com> Tested-by: Andre Przywara <Andre.Przywara@arm.com> Fixes: 32f139551954 ("arm/arm64: KVM: Statically configure the host's view of MPIDR") Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Diffstat (limited to 'tools/testing/selftests')
0 files changed, 0 insertions, 0 deletions