@@ -10,7 +10,7 @@ typedef struct {
#else
int switch_pending;
#endif
- unsigned int vmalloc_seq;
+ atomic_t vmalloc_seq;
unsigned long sigpage;
#ifdef CONFIG_VDSO
unsigned long vdso;
@@ -23,6 +23,16 @@
void __check_vmalloc_seq(struct mm_struct *mm);
+#ifdef CONFIG_MMU
+static inline void check_vmalloc_seq(struct mm_struct *mm)
+{
+ if (!IS_ENABLED(CONFIG_ARM_LPAE) &&
+ unlikely(atomic_read(&mm->context.vmalloc_seq) !=
+ atomic_read(&init_mm.context.vmalloc_seq)))
+ __check_vmalloc_seq(mm);
+}
+#endif
+
#ifdef CONFIG_CPU_HAS_ASID
void check_and_switch_context(struct mm_struct *mm, struct task_struct *tsk);
@@ -52,8 +62,7 @@ static inline void a15_erratum_get_cpumask(int this_cpu, struct mm_struct *mm,
static inline void check_and_switch_context(struct mm_struct *mm,
struct task_struct *tsk)
{
- if (unlikely(mm->context.vmalloc_seq != init_mm.context.vmalloc_seq))
- __check_vmalloc_seq(mm);
+ check_vmalloc_seq(mm);
if (irqs_disabled())
/*
@@ -240,8 +240,7 @@ void check_and_switch_context(struct mm_struct *mm, struct task_struct *tsk)
unsigned int cpu = smp_processor_id();
u64 asid;
- if (unlikely(mm->context.vmalloc_seq != init_mm.context.vmalloc_seq))
- __check_vmalloc_seq(mm);
+ check_vmalloc_seq(mm);
/*
* We cannot update the pgd and the ASID atomicly with classic
@@ -117,16 +117,21 @@ EXPORT_SYMBOL(ioremap_page);
void __check_vmalloc_seq(struct mm_struct *mm)
{
- unsigned int seq;
+ int seq;
do {
- seq = init_mm.context.vmalloc_seq;
+ seq = atomic_read(&init_mm.context.vmalloc_seq);
memcpy(pgd_offset(mm, VMALLOC_START),
pgd_offset_k(VMALLOC_START),
sizeof(pgd_t) * (pgd_index(VMALLOC_END) -
pgd_index(VMALLOC_START)));
- mm->context.vmalloc_seq = seq;
- } while (seq != init_mm.context.vmalloc_seq);
+ /*
+ * Use a store-release so that other CPUs that observe the
+ * counter's new value are guaranteed to see the results of the
+ * memcpy as well.
+ */
+ atomic_set_release(&mm->context.vmalloc_seq, seq);
+ } while (seq != atomic_read(&init_mm.context.vmalloc_seq));
}
#if !defined(CONFIG_SMP) && !defined(CONFIG_ARM_LPAE)
@@ -157,7 +162,7 @@ static void unmap_area_sections(unsigned long virt, unsigned long size)
* Note: this is still racy on SMP machines.
*/
pmd_clear(pmdp);
- init_mm.context.vmalloc_seq++;
+ atomic_inc_return_release(&init_mm.context.vmalloc_seq);
/*
* Free the page table, if there was one.
@@ -174,8 +179,7 @@ static void unmap_area_sections(unsigned long virt, unsigned long size)
* Ensure that the active_mm is up to date - we want to
* catch any use-after-iounmap cases.
*/
- if (current->active_mm->context.vmalloc_seq != init_mm.context.vmalloc_seq)
- __check_vmalloc_seq(current->active_mm);
+ check_vmalloc_seq(current->active_mm);
flush_tlb_kernel_range(virt, end);
}
Currently, the vmalloc_seq counter is only used to keep track of changes in the vmalloc region on !SMP builds, which means there is no need to deal with concurrency explicitly. However, in a subsequent patch, we will wire up this same mechanism for ensuring that vmap'ed stacks are guaranteed to be mapped by the active mm before switching to a task, and here we need to ensure that changes to the page tables are visible to other CPUs when they observe a change in the sequence count. Since LPAE needs none of this, fold a check against it into the vmalloc_seq counter check after breaking it out into a separate static inline helper. Signed-off-by: Ard Biesheuvel <ardb@kernel.org> --- arch/arm/include/asm/mmu.h | 2 +- arch/arm/include/asm/mmu_context.h | 13 +++++++++++-- arch/arm/mm/context.c | 3 +-- arch/arm/mm/ioremap.c | 18 +++++++++++------- 4 files changed, 24 insertions(+), 12 deletions(-)