| Message ID | 20210922135924.3109291-1-dann.frazier@canonical.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | arm64: Restore forced disabling of KPTI on ThunderX | expand |
On Wed, Sep 22, 2021 at 07:59:24AM -0600, dann frazier wrote: > A noted side-effect of commit 0c6c2d3615ef ("arm64: Generate cpucaps.h") > is that cpucaps are now sorted, changing the enumeration order. This > assumed no dependencies between cpucaps, which turned out not to be true Reviwed-by: Mark Brown <broonie@kernel.org>
On Wed, 22 Sep 2021 14:59:24 +0100, dann frazier <dann.frazier@canonical.com> wrote: > > A noted side-effect of commit 0c6c2d3615ef ("arm64: Generate cpucaps.h") > is that cpucaps are now sorted, changing the enumeration order. This > assumed no dependencies between cpucaps, which turned out not to be true > in one case. UNMAP_KERNEL_AT_EL0 currently needs to be processed after > WORKAROUND_CAVIUM_27456. ThunderX systems are incompatible with KPTI, so > unmap_kernel_at_el0() bails if WORKAROUND_CAVIUM_27456 is set. But because > of the sorting, WORKAROUND_CAVIUM_27456 will not yet have been considered > when unmap_kernel_at_el0() checks for it, so the kernel tries to > run w/ KPTI - and quickly falls over. > > Because all ThunderX implementations have homogeneous CPUs, we can remove > this dependency by just checking the current CPU for the erratum. > > Fixes: 0c6c2d3615ef ("arm64: Generate cpucaps.h") > Suggested-by: Suzuki K Poulose <suzuki.poulose@arm.com> > Cc: stable@vger.kernel.org # 5.13+ > Signed-off-by: dann frazier <dann.frazier@canonical.com> > --- > arch/arm64/kernel/cpufeature.c | 2 +- > 1 file changed, 1 insertion(+), 1 deletion(-) > > diff --git a/arch/arm64/kernel/cpufeature.c b/arch/arm64/kernel/cpufeature.c > index f8a3067d10c6..7275b49034f3 100644 > --- a/arch/arm64/kernel/cpufeature.c > +++ b/arch/arm64/kernel/cpufeature.c > @@ -1528,7 +1528,7 @@ static bool unmap_kernel_at_el0(const struct arm64_cpu_capabilities *entry, > * ThunderX leads to apparent I-cache corruption of kernel text, which > * ends as well as you might imagine. Don't even try. > */ > - if (cpus_have_const_cap(ARM64_WORKAROUND_CAVIUM_27456)) { > + if (this_cpu_has_cap(ARM64_WORKAROUND_CAVIUM_27456)) { > str = "ARM64_WORKAROUND_CAVIUM_27456"; > __kpti_forced = -1; > } Ouch, nice catch. Hopefully, nobody will build a big-little system using TX1 in this instance of the universe. Acked-by: Marc Zyngier <maz@kernel.org> M.
On 22/09/2021 14:59, dann frazier wrote: > A noted side-effect of commit 0c6c2d3615ef ("arm64: Generate cpucaps.h") > is that cpucaps are now sorted, changing the enumeration order. This > assumed no dependencies between cpucaps, which turned out not to be true > in one case. UNMAP_KERNEL_AT_EL0 currently needs to be processed after > WORKAROUND_CAVIUM_27456. ThunderX systems are incompatible with KPTI, so > unmap_kernel_at_el0() bails if WORKAROUND_CAVIUM_27456 is set. But because > of the sorting, WORKAROUND_CAVIUM_27456 will not yet have been considered > when unmap_kernel_at_el0() checks for it, so the kernel tries to > run w/ KPTI - and quickly falls over. > > Because all ThunderX implementations have homogeneous CPUs, we can remove > this dependency by just checking the current CPU for the erratum. > > Fixes: 0c6c2d3615ef ("arm64: Generate cpucaps.h") > Suggested-by: Suzuki K Poulose <suzuki.poulose@arm.com> > Cc: stable@vger.kernel.org # 5.13+ > Signed-off-by: dann frazier <dann.frazier@canonical.com> > --- > arch/arm64/kernel/cpufeature.c | 2 +- > 1 file changed, 1 insertion(+), 1 deletion(-) > > diff --git a/arch/arm64/kernel/cpufeature.c b/arch/arm64/kernel/cpufeature.c > index f8a3067d10c6..7275b49034f3 100644 > --- a/arch/arm64/kernel/cpufeature.c > +++ b/arch/arm64/kernel/cpufeature.c > @@ -1528,7 +1528,7 @@ static bool unmap_kernel_at_el0(const struct arm64_cpu_capabilities *entry, > * ThunderX leads to apparent I-cache corruption of kernel text, which > * ends as well as you might imagine. Don't even try. > */ > - if (cpus_have_const_cap(ARM64_WORKAROUND_CAVIUM_27456)) { > + if (this_cpu_has_cap(ARM64_WORKAROUND_CAVIUM_27456)) { Please could you also update the comment right above this line to explain, why we do this and why this is fine (just like you have in the description) ? Something like : * Since we cannot rely on the order of the cpucaps * we cannot rely on the cpus_have_*cap() helpers to * detect the erratum on the system. However, since * affected CPUs are always in a homoegeneous configuration * we could rely on this_cpu_has_cap() */ So that looking at the code, it is easier to comprehend what we figured out in the mailing list (and the description) With that: Reviewed-by: Suzuki K Poulose <suzuki.poulose@arm.com> Suzuki
On Thu, Sep 23, 2021 at 10:41:00AM +0100, Suzuki K Poulose wrote: > On 22/09/2021 14:59, dann frazier wrote: > > A noted side-effect of commit 0c6c2d3615ef ("arm64: Generate cpucaps.h") > > is that cpucaps are now sorted, changing the enumeration order. This > > assumed no dependencies between cpucaps, which turned out not to be true > > in one case. UNMAP_KERNEL_AT_EL0 currently needs to be processed after > > WORKAROUND_CAVIUM_27456. ThunderX systems are incompatible with KPTI, so > > unmap_kernel_at_el0() bails if WORKAROUND_CAVIUM_27456 is set. But because > > of the sorting, WORKAROUND_CAVIUM_27456 will not yet have been considered > > when unmap_kernel_at_el0() checks for it, so the kernel tries to > > run w/ KPTI - and quickly falls over. > > > > Because all ThunderX implementations have homogeneous CPUs, we can remove > > this dependency by just checking the current CPU for the erratum. > > > > Fixes: 0c6c2d3615ef ("arm64: Generate cpucaps.h") > > Suggested-by: Suzuki K Poulose <suzuki.poulose@arm.com> > > Cc: stable@vger.kernel.org # 5.13+ > > Signed-off-by: dann frazier <dann.frazier@canonical.com> > > --- > > arch/arm64/kernel/cpufeature.c | 2 +- > > 1 file changed, 1 insertion(+), 1 deletion(-) > > > > diff --git a/arch/arm64/kernel/cpufeature.c b/arch/arm64/kernel/cpufeature.c > > index f8a3067d10c6..7275b49034f3 100644 > > --- a/arch/arm64/kernel/cpufeature.c > > +++ b/arch/arm64/kernel/cpufeature.c > > @@ -1528,7 +1528,7 @@ static bool unmap_kernel_at_el0(const struct arm64_cpu_capabilities *entry, > > * ThunderX leads to apparent I-cache corruption of kernel text, which > > * ends as well as you might imagine. Don't even try. > > */ > > - if (cpus_have_const_cap(ARM64_WORKAROUND_CAVIUM_27456)) { > > + if (this_cpu_has_cap(ARM64_WORKAROUND_CAVIUM_27456)) { > > Please could you also update the comment right above this line to > explain, why we do this and why this is fine (just like you have > in the description) ? Something like : > > * Since we cannot rely on the order of the cpucaps > * we cannot rely on the cpus_have_*cap() helpers to > * detect the erratum on the system. However, since > * affected CPUs are always in a homoegeneous configuration > * we could rely on this_cpu_has_cap() > */ > > So that looking at the code, it is easier to comprehend what we > figured out in the mailing list (and the description) Sure thing, v2 coming shortly. -dann > With that: > > Reviewed-by: Suzuki K Poulose <suzuki.poulose@arm.com> > > Suzuki
diff --git a/arch/arm64/kernel/cpufeature.c b/arch/arm64/kernel/cpufeature.c index f8a3067d10c6..7275b49034f3 100644 --- a/arch/arm64/kernel/cpufeature.c +++ b/arch/arm64/kernel/cpufeature.c @@ -1528,7 +1528,7 @@ static bool unmap_kernel_at_el0(const struct arm64_cpu_capabilities *entry, * ThunderX leads to apparent I-cache corruption of kernel text, which * ends as well as you might imagine. Don't even try. */ - if (cpus_have_const_cap(ARM64_WORKAROUND_CAVIUM_27456)) { + if (this_cpu_has_cap(ARM64_WORKAROUND_CAVIUM_27456)) { str = "ARM64_WORKAROUND_CAVIUM_27456"; __kpti_forced = -1; }
A noted side-effect of commit 0c6c2d3615ef ("arm64: Generate cpucaps.h") is that cpucaps are now sorted, changing the enumeration order. This assumed no dependencies between cpucaps, which turned out not to be true in one case. UNMAP_KERNEL_AT_EL0 currently needs to be processed after WORKAROUND_CAVIUM_27456. ThunderX systems are incompatible with KPTI, so unmap_kernel_at_el0() bails if WORKAROUND_CAVIUM_27456 is set. But because of the sorting, WORKAROUND_CAVIUM_27456 will not yet have been considered when unmap_kernel_at_el0() checks for it, so the kernel tries to run w/ KPTI - and quickly falls over. Because all ThunderX implementations have homogeneous CPUs, we can remove this dependency by just checking the current CPU for the erratum. Fixes: 0c6c2d3615ef ("arm64: Generate cpucaps.h") Suggested-by: Suzuki K Poulose <suzuki.poulose@arm.com> Cc: stable@vger.kernel.org # 5.13+ Signed-off-by: dann frazier <dann.frazier@canonical.com> --- arch/arm64/kernel/cpufeature.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)