On 07/11/2019 14:37, Mark Brown wrote:

On Thu, Nov 07, 2019 at 12:01:10PM +0000, Suzuki K Poulose wrote:

quoted

On 06/11/2019 13:00, Mark Brown wrote:

quoted

quoted

+     /*
+      * E0PD does a similar job to KPTI so can be used instead
+      * where available.
+      */
+     if (IS_ENABLED(CONFIG_ARM64_E0PD)) {
+             ftr = read_sysreg_s(SYS_ID_AA64MMFR2_EL1);

quoted

I am trying to write down the rationale of checking this per-CPU.

quoted

Given that this gets run on all individual CPUs, via unmap_kernel_at_el0()
and the decision of choosing KPTI is affected by the lack of the E0PD feature
when it is helpful, having CPU local check is fine. Also this gives us the
advantage of choosing an nG mapping when the boot CPU indicates the need.

Well, it's mainly the fact that this runs really early on in boot before
the cpufeature code has fully initialized so as with the existing code
immediately below for identifying TX1 we can't rely on the cpufeature
code being done.

Yes, I acknowledge that. I was writing it down to clear why this was
fine and why it has its own advantage. This may not be obvious for
someone who reads it later. So having this in a comment helps to
avoid staring at it.

Suzuki

_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel