ARM: mm: make vmalloc_seq handling SMP safe

Rework the vmalloc_seq handling so it can be used safely under SMP, as
we started using it to ensure 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.

Given that vmap'ed stacks are now also supported on !SMP configurations,
let's drop the WARN() that could potentially now fire spuriously.

Signed-off-by: Ard Biesheuvel <ardb@kernel.org>

diff --git a/arch/arm/include/asm/mmu.h b/arch/arm/include/asm/mmu.h
index 1592a4264488e2c0646c5500fc6e4b8c44d5247a..e049723840d3298a42f6b3c39b9ac558fee1857d 100644 (file)
--- a/arch/arm/include/asm/mmu.h
+++ b/arch/arm/include/asm/mmu.h
@@ -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;

diff --git a/arch/arm/include/asm/mmu_context.h b/arch/arm/include/asm/mmu_context.h
index 84e58956fcab9ac32c8562bf1e3837b92d3efedd..db2cb06aa8cf5e9d5253a33b84d31b1a8a6ab828 100644 (file)
--- a/arch/arm/include/asm/mmu_context.h
+++ b/arch/arm/include/asm/mmu_context.h
@@ -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())
                /*
@@ -129,6 +138,15 @@ switch_mm(struct mm_struct *prev, struct mm_struct *next,
 #endif
 }
 
+#ifdef CONFIG_VMAP_STACK
+static inline void enter_lazy_tlb(struct mm_struct *mm, struct task_struct *tsk)
+{
+       if (mm != &init_mm)
+               check_vmalloc_seq(mm);
+}
+#define enter_lazy_tlb enter_lazy_tlb
+#endif
+
 #include <asm-generic/mmu_context.h>
 
 #endif

diff --git a/arch/arm/include/asm/page.h b/arch/arm/include/asm/page.h
index 7b871ed99ccf0a63a5b883913997abf78adcff91..5fcc8a600e36d4c62ad53faabc7c7d3df4d92527 100644 (file)
--- a/arch/arm/include/asm/page.h
+++ b/arch/arm/include/asm/page.h
@@ -147,11 +147,10 @@ extern void copy_page(void *to, const void *from);
 #include <asm/pgtable-3level-types.h>
 #else
 #include <asm/pgtable-2level-types.h>
-#endif
-
 #ifdef CONFIG_VMAP_STACK
 #define ARCH_PAGE_TABLE_SYNC_MASK      PGTBL_PMD_MODIFIED
 #endif
+#endif
 
 #endif /* CONFIG_MMU */
 

diff --git a/arch/arm/kernel/traps.c b/arch/arm/kernel/traps.c
index 3f38357efc4627c585c8a14f52db40a85387be46..08612032aefee7249ad000caa2dc9ef4041d5a13 100644 (file)
--- a/arch/arm/kernel/traps.c
+++ b/arch/arm/kernel/traps.c
@@ -885,6 +885,7 @@ asmlinkage void handle_bad_stack(struct pt_regs *regs)
        die("kernel stack overflow", regs, 0);
 }
 
+#ifndef CONFIG_ARM_LPAE
 /*
  * Normally, we rely on the logic in do_translation_fault() to update stale PMD
  * entries covering the vmalloc space in a task's page tables when it first
@@ -895,26 +896,14 @@ asmlinkage void handle_bad_stack(struct pt_regs *regs)
  * So we need to ensure that these PMD entries are up to date *before* the MM
  * switch. As we already have some logic in the MM switch path that takes care
  * of this, let's trigger it by bumping the counter every time the core vmalloc
- * code modifies a PMD entry in the vmalloc region.
+ * code modifies a PMD entry in the vmalloc region. Use release semantics on
+ * the store so that other CPUs observing the counter's new value are
+ * guaranteed to see the updated page table entries as well.
  */
 void arch_sync_kernel_mappings(unsigned long start, unsigned long end)
 {
-       if (start > VMALLOC_END || end < VMALLOC_START)
-               return;
-
-       /*
-        * This hooks into the core vmalloc code to receive notifications of
-        * any PMD level changes that have been made to the kernel page tables.
-        * This means it should only be triggered once for every MiB worth of
-        * vmalloc space, given that we don't support huge vmalloc/vmap on ARM,
-        * and that kernel PMD level table entries are rarely (if ever)
-        * updated.
-        *
-        * This means that the counter is going to max out at ~250 for the
-        * typical case. If it overflows, something entirely unexpected has
-        * occurred so let's throw a warning if that happens.
-        */
-       WARN_ON(++init_mm.context.vmalloc_seq == UINT_MAX);
+       if (start < VMALLOC_END && end > VMALLOC_START)
+               atomic_inc_return_release(&init_mm.context.vmalloc_seq);
 }
-
+#endif
 #endif

diff --git a/arch/arm/mm/context.c b/arch/arm/mm/context.c
index 48091870db89e420da8a9fa193420a9ac2e991f0..4204ffa2d104f144fc06e326d8005b83caaf09a2 100644 (file)
--- a/arch/arm/mm/context.c
+++ b/arch/arm/mm/context.c
@@ -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

diff --git a/arch/arm/mm/ioremap.c b/arch/arm/mm/ioremap.c
index 6e830b9418c94c7595c51f4cc4abb951d7c3ed29..8963c8c6347175295e9b78b51524ca7079a04ef8 100644 (file)
--- a/arch/arm/mm/ioremap.c
+++ b/arch/arm/mm/ioremap.c
@@ -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);
 }