On Thu, Dec 06, 2018 at 03:30:22PM +0100, Krzysztof Kozlowski wrote:

quoted

On Thu, 6 Dec 2018 at 15:07, Russell King - ARM Linux

quoted

That basically means that the dcache_clean_area method for CPU1
differs from the dcache_clean_area method for CPU0.  If all your
CPUs are identical, that definitely should not be happening.

Hmm.  Interestingly, OMAP4430 passes hotplug tests just fine.

Please try this patch.

This fixes both hotplug and suspend to RAM. I was trying to narrow why
the pointers to all processor functions differ. During first boot they
were OK but it seems they were changed just before suspend.

Thanks for testing.  I think this is probably a better patch which
should end up with the same result.

I suspect no one else has noticed because most people have big.Little
support disabled - that'd explain why it doesn't show up on OMAP4.

diff --git a/arch/arm/mm/proc-macros.S b/arch/arm/mm/proc-macros.S
index 81d0efb055c6..44f9776139a8 100644
--- a/arch/arm/mm/proc-macros.S
+++ b/arch/arm/mm/proc-macros.S

@@ -274,6 +274,13 @@
        .endm

 .macro define_processor_functions name:req, dabort:req, pabort:req, nommu=0, suspend=0, bugs=0
+/*
+ * If we are building for big.Little with branch predictor hardening,
+ * we need the processor function tables to remain available after boot.
+ */
+#if defined(CONFIG_BIG_LITTLE) && defined(CONFIG_HARDEN_BRANCH_PREDICTOR)
+       .rodata
+#endif
        .type   \name\()_processor_functions, #object
        .align 2
 ENTRY(\name\()_processor_functions)

@@ -309,6 +316,9 @@ ENTRY(\name\()_processor_functions)
        .endif

        .size   \name\()_processor_functions, . - \name\()_processor_functions
+#if defined(CONFIG_BIG_LITTLE) && defined(CONFIG_HARDEN_BRANCH_PREDICTOR)
+       .previous
+#endif
 .endm

 .macro define_cache_functions name:req