ARM: 7684/1: errata: Workaround for Cortex-A15 erratum 798181 (TLBI/DSB operations)

On Cortex-A15 (r0p0..r3p2) the TLBI/DSB are not adequately shooting down
all use of the old entries. This patch implements the erratum workaround
which consists of:

1. Dummy TLBIMVAIS and DSB on the CPU doing the TLBI operation.
2. Send IPI to the CPUs that are running the same mm (and ASID) as the
   one being invalidated (or all the online CPUs for global pages).
3. CPU receiving the IPI executes a DMB and CLREX (part of the exception
   return code already).

Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>

arch/arm/Kconfig

@@ -1439,6 +1439,16 @@ config ARM_ERRATA_775420
 	 to deadlock. This workaround puts DSB before executing ISB if
 	 an abort may occur on cache maintenance.
 
+config ARM_ERRATA_798181
+	bool "ARM errata: TLBI/DSB failure on Cortex-A15"
+	depends on CPU_V7 && SMP
+	help
+	  On Cortex-A15 (r0p0..r3p2) the TLBI*IS/DSB operations are not
+	  adequately shooting down all use of the old entries. This
+	  option enables the Linux kernel workaround for this erratum
+	  which sends an IPI to the CPUs that are running the same ASID
+	  as the one being invalidated.
+
 endmenu
 
 source "arch/arm/common/Kconfig"

arch/arm/include/asm/highmem.h

@@ -41,6 +41,13 @@ extern void kunmap_high(struct page *page);
 #endif
 #endif
 
+/*
+ * Needed to be able to broadcast the TLB invalidation for kmap.
+ */
+#ifdef CONFIG_ARM_ERRATA_798181
+#undef ARCH_NEEDS_KMAP_HIGH_GET
+#endif
+
 #ifdef ARCH_NEEDS_KMAP_HIGH_GET
 extern void *kmap_high_get(struct page *page);
 #else

arch/arm/include/asm/mmu_context.h

@@ -27,6 +27,8 @@ void __check_vmalloc_seq(struct mm_struct *mm);
 void check_and_switch_context(struct mm_struct *mm, struct task_struct *tsk);
 #define init_new_context(tsk,mm)	({ atomic64_set(&mm->context.id, 0); 0; })
 
+DECLARE_PER_CPU(atomic64_t, active_asids);
+
 #else	/* !CONFIG_CPU_HAS_ASID */
 
 #ifdef CONFIG_MMU

arch/arm/include/asm/tlbflush.h

@@ -450,6 +450,21 @@ static inline void local_flush_bp_all(void)
 		isb();
 }
 
+#ifdef CONFIG_ARM_ERRATA_798181
+static inline void dummy_flush_tlb_a15_erratum(void)
+{
+	/*
+	 * Dummy TLBIMVAIS. Using the unmapped address 0 and ASID 0.
+	 */
+	asm("mcr p15, 0, %0, c8, c3, 1" : : "r" (0));
+	dsb();
+}
+#else
+static inline void dummy_flush_tlb_a15_erratum(void)
+{
+}
+#endif
+
 /*
  *	flush_pmd_entry
  *

arch/arm/kernel/smp_tlb.c

@@ -12,6 +12,7 @@
 
 #include <asm/smp_plat.h>
 #include <asm/tlbflush.h>
+#include <asm/mmu_context.h>
 
 /**********************************************************************/
 
@@ -69,12 +70,72 @@ static inline void ipi_flush_bp_all(void *ignored)
 	local_flush_bp_all();
 }
 
+#ifdef CONFIG_ARM_ERRATA_798181
+static int erratum_a15_798181(void)
+{
+	unsigned int midr = read_cpuid_id();
+
+	/* Cortex-A15 r0p0..r3p2 affected */
+	if ((midr & 0xff0ffff0) != 0x410fc0f0 || midr > 0x413fc0f2)
+		return 0;
+	return 1;
+}
+#else
+static int erratum_a15_798181(void)
+{
+	return 0;
+}
+#endif
+
+static void ipi_flush_tlb_a15_erratum(void *arg)
+{
+	dmb();
+}
+
+static void broadcast_tlb_a15_erratum(void)
+{
+	if (!erratum_a15_798181())
+		return;
+
+	dummy_flush_tlb_a15_erratum();
+	smp_call_function_many(cpu_online_mask, ipi_flush_tlb_a15_erratum,
+			       NULL, 1);
+}
+
+static void broadcast_tlb_mm_a15_erratum(struct mm_struct *mm)
+{
+	int cpu;
+	cpumask_t mask = { CPU_BITS_NONE };
+
+	if (!erratum_a15_798181())
+		return;
+
+	dummy_flush_tlb_a15_erratum();
+	for_each_online_cpu(cpu) {
+		if (cpu == smp_processor_id())
+			continue;
+		/*
+		 * We only need to send an IPI if the other CPUs are running
+		 * the same ASID as the one being invalidated. There is no
+		 * need for locking around the active_asids check since the
+		 * switch_mm() function has at least one dmb() (as required by
+		 * this workaround) in case a context switch happens on
+		 * another CPU after the condition below.
+		 */
+		if (atomic64_read(&mm->context.id) ==
+		    atomic64_read(&per_cpu(active_asids, cpu)))
+			cpumask_set_cpu(cpu, &mask);
+	}
+	smp_call_function_many(&mask, ipi_flush_tlb_a15_erratum, NULL, 1);
+}
+
 void flush_tlb_all(void)
 {
 	if (tlb_ops_need_broadcast())
 		on_each_cpu(ipi_flush_tlb_all, NULL, 1);
 	else
 		local_flush_tlb_all();
+	broadcast_tlb_a15_erratum();
 }
 
 void flush_tlb_mm(struct mm_struct *mm)
@@ -83,6 +144,7 @@ void flush_tlb_mm(struct mm_struct *mm)
 		on_each_cpu_mask(mm_cpumask(mm), ipi_flush_tlb_mm, mm, 1);
 	else
 		local_flush_tlb_mm(mm);
+	broadcast_tlb_mm_a15_erratum(mm);
 }
 
 void flush_tlb_page(struct vm_area_struct *vma, unsigned long uaddr)
@@ -95,6 +157,7 @@ void flush_tlb_page(struct vm_area_struct *vma, unsigned long uaddr)
 					&ta, 1);
 	} else
 		local_flush_tlb_page(vma, uaddr);
+	broadcast_tlb_mm_a15_erratum(vma->vm_mm);
 }
 
 void flush_tlb_kernel_page(unsigned long kaddr)
@@ -105,6 +168,7 @@ void flush_tlb_kernel_page(unsigned long kaddr)
 		on_each_cpu(ipi_flush_tlb_kernel_page, &ta, 1);
 	} else
 		local_flush_tlb_kernel_page(kaddr);
+	broadcast_tlb_a15_erratum();
 }
 
 void flush_tlb_range(struct vm_area_struct *vma,
@@ -119,6 +183,7 @@ void flush_tlb_range(struct vm_area_struct *vma,
 					&ta, 1);
 	} else
 		local_flush_tlb_range(vma, start, end);
+	broadcast_tlb_mm_a15_erratum(vma->vm_mm);
 }
 
 void flush_tlb_kernel_range(unsigned long start, unsigned long end)
@@ -130,6 +195,7 @@ void flush_tlb_kernel_range(unsigned long start, unsigned long end)
 		on_each_cpu(ipi_flush_tlb_kernel_range, &ta, 1);
 	} else
 		local_flush_tlb_kernel_range(start, end);
+	broadcast_tlb_a15_erratum();
 }
 
 void flush_bp_all(void)

arch/arm/mm/context.c

@@ -48,7 +48,7 @@ static DEFINE_RAW_SPINLOCK(cpu_asid_lock);
 static atomic64_t asid_generation = ATOMIC64_INIT(ASID_FIRST_VERSION);
 static DECLARE_BITMAP(asid_map, NUM_USER_ASIDS);
 
-static DEFINE_PER_CPU(atomic64_t, active_asids);
+DEFINE_PER_CPU(atomic64_t, active_asids);
 static DEFINE_PER_CPU(u64, reserved_asids);
 static cpumask_t tlb_flush_pending;
 
@@ -215,6 +215,7 @@ void check_and_switch_context(struct mm_struct *mm, struct task_struct *tsk)
 	if (cpumask_test_and_clear_cpu(cpu, &tlb_flush_pending)) {
 		local_flush_bp_all();
 		local_flush_tlb_all();
+		dummy_flush_tlb_a15_erratum();
 	}
 
 	atomic64_set(&per_cpu(active_asids, cpu), asid);