xtensa: check TLB sanity on return to userspace

- check that user TLB mappings correspond to the current page table;
- check that TLB mapping VPN is in the kernel/user address range
  in accordance with its ASID.

Signed-off-by: Max Filippov <jcmvbkbc@gmail.com>
Signed-off-by: Chris Zankel <chris@zankel.net>

3 files changed, 131 insertions, 1 deletions

diff --git a/arch/xtensa/Kconfig.debug b/arch/xtensa/Kconfig.debug
index a34010e0e51c..af7da74d535f 100644
--- a/arch/xtensa/Kconfig.debug
+++ b/arch/xtensa/Kconfig.debug
@@ -2,6 +2,16 @@ menu "Kernel hacking"
 
 source "lib/Kconfig.debug"
 
+config DEBUG_TLB_SANITY
+	bool "Debug TLB sanity"
+	depends on DEBUG_KERNEL
+	help
+	  Enable this to turn on TLB sanity check on each entry to userspace.
+	  This check can spot missing TLB invalidation/wrong PTE permissions/
+	  premature page freeing.
+
+	  If unsure, say N.
+
 config LD_NO_RELAX
 	bool "Disable linker relaxation"
 	default n
diff --git a/arch/xtensa/kernel/entry.S b/arch/xtensa/kernel/entry.S
index fa94512ff84d..9298742f0fd0 100644
--- a/arch/xtensa/kernel/entry.S
+++ b/arch/xtensa/kernel/entry.S
@@ -458,7 +458,7 @@ common_exception_return:
 
 	_bbsi.l	a4, TIF_NEED_RESCHED, 3f
 	_bbsi.l	a4, TIF_NOTIFY_RESUME, 2f
-	_bbci.l	a4, TIF_SIGPENDING, 4f
+	_bbci.l	a4, TIF_SIGPENDING, 5f
 
 2:	l32i	a4, a1, PT_DEPC
 	bgeui	a4, VALID_DOUBLE_EXCEPTION_ADDRESS, 4f
@@ -476,6 +476,13 @@ common_exception_return:
 	callx4	a4
 	j	1b
 
+5:
+#ifdef CONFIG_DEBUG_TLB_SANITY
+	l32i	a4, a1, PT_DEPC
+	bgeui	a4, VALID_DOUBLE_EXCEPTION_ADDRESS, 4f
+	movi	a4, check_tlb_sanity
+	callx4	a4
+#endif
 4:	/* Restore optional registers. */
 
 	load_xtregs_opt a1 a2 a4 a5 a6 a7 PT_XTREGS_OPT
diff --git a/arch/xtensa/mm/tlb.c b/arch/xtensa/mm/tlb.c
index 743346150eea..ca9d2366bf12 100644
--- a/arch/xtensa/mm/tlb.c
+++ b/arch/xtensa/mm/tlb.c
@@ -141,3 +141,116 @@ void flush_tlb_page (struct vm_area_struct *vma, unsigned long page)
 
 	local_irq_restore(flags);
 }
+
+#ifdef CONFIG_DEBUG_TLB_SANITY
+
+static unsigned get_pte_for_vaddr(unsigned vaddr)
+{
+	struct task_struct *task = get_current();
+	struct mm_struct *mm = task->mm;
+	pgd_t *pgd;
+	pmd_t *pmd;
+	pte_t *pte;
+
+	if (!mm)
+		mm = task->active_mm;
+	pgd = pgd_offset(mm, vaddr);
+	if (pgd_none_or_clear_bad(pgd))
+		return 0;
+	pmd = pmd_offset(pgd, vaddr);
+	if (pmd_none_or_clear_bad(pmd))
+		return 0;
+	pte = pte_offset_map(pmd, vaddr);
+	if (!pte)
+		return 0;
+	return pte_val(*pte);
+}
+
+enum {
+	TLB_SUSPICIOUS	= 1,
+	TLB_INSANE	= 2,
+};
+
+static void tlb_insane(void)
+{
+	BUG_ON(1);
+}
+
+static void tlb_suspicious(void)
+{
+	WARN_ON(1);
+}
+
+/*
+ * Check that TLB entries with kernel ASID (1) have kernel VMA (>= TASK_SIZE),
+ * and TLB entries with user ASID (>=4) have VMA < TASK_SIZE.
+ *
+ * Check that valid TLB entries either have the same PA as the PTE, or PTE is
+ * marked as non-present. Non-present PTE and the page with non-zero refcount
+ * and zero mapcount is normal for batched TLB flush operation. Zero refcount
+ * means that the page was freed prematurely. Non-zero mapcount is unusual,
+ * but does not necessary means an error, thus marked as suspicious.
+ */
+static int check_tlb_entry(unsigned w, unsigned e, bool dtlb)
+{
+	unsigned tlbidx = w | (e << PAGE_SHIFT);
+	unsigned r0 = dtlb ?
+		read_dtlb_virtual(tlbidx) : read_itlb_virtual(tlbidx);
+	unsigned vpn = (r0 & PAGE_MASK) | (e << PAGE_SHIFT);
+	unsigned pte = get_pte_for_vaddr(vpn);
+	unsigned mm_asid = (get_rasid_register() >> 8) & ASID_MASK;
+	unsigned tlb_asid = r0 & ASID_MASK;
+	bool kernel = tlb_asid == 1;
+	int rc = 0;
+
+	if (tlb_asid > 0 && ((vpn < TASK_SIZE) == kernel)) {
+		pr_err("%cTLB: way: %u, entry: %u, VPN %08x in %s PTE\n",
+				dtlb ? 'D' : 'I', w, e, vpn,
+				kernel ? "kernel" : "user");
+		rc |= TLB_INSANE;
+	}
+
+	if (tlb_asid == mm_asid) {
+		unsigned r1 = dtlb ? read_dtlb_translation(tlbidx) :
+			read_itlb_translation(tlbidx);
+		if ((pte ^ r1) & PAGE_MASK) {
+			pr_err("%cTLB: way: %u, entry: %u, mapping: %08x->%08x, PTE: %08x\n",
+					dtlb ? 'D' : 'I', w, e, r0, r1, pte);
+			if (pte == 0 || !pte_present(__pte(pte))) {
+				struct page *p = pfn_to_page(r1 >> PAGE_SHIFT);
+				pr_err("page refcount: %d, mapcount: %d\n",
+						page_count(p),
+						page_mapcount(p));
+				if (!page_count(p))
+					rc |= TLB_INSANE;
+				else if (page_mapped(p))
+					rc |= TLB_SUSPICIOUS;
+			} else {
+				rc |= TLB_INSANE;
+			}
+		}
+	}
+	return rc;
+}
+
+void check_tlb_sanity(void)
+{
+	unsigned long flags;
+	unsigned w, e;
+	int bug = 0;
+
+	local_irq_save(flags);
+	for (w = 0; w < DTLB_ARF_WAYS; ++w)
+		for (e = 0; e < (1 << XCHAL_DTLB_ARF_ENTRIES_LOG2); ++e)
+			bug |= check_tlb_entry(w, e, true);
+	for (w = 0; w < ITLB_ARF_WAYS; ++w)
+		for (e = 0; e < (1 << XCHAL_ITLB_ARF_ENTRIES_LOG2); ++e)
+			bug |= check_tlb_entry(w, e, false);
+	if (bug & TLB_INSANE)
+		tlb_insane();
+	if (bug & TLB_SUSPICIOUS)
+		tlb_suspicious();
+	local_irq_restore(flags);
+}
+
+#endif /* CONFIG_DEBUG_TLB_SANITY */