Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
From:   Maninder Singh <maninder1.s@samsung.com>
To:     dave.hansen@linux.intel.com, luto@kernel.org, peterz@infradead.org,
        tglx@linutronix.de, mingo@redhat.com, bp@alien8.de, hpa@zytor.com
Cc:     x86@kernel.org, linux-kernel@vger.kernel.org, v.narang@samsung.com,
        a.sahrawat@samsung.com, avnish.jain@samsung.com,
        Maninder Singh <maninder1.s@samsung.com>
Subject: [PATCH 1/1] mm: Map next task stack in previous mm.
Date:   Tue, 19 Nov 2019 10:45:20 +0530
Message-Id: <1574140520-9738-1-git-send-email-maninder1.s@samsung.com>
Content-Type: text/plain; charset="utf-8"
CMS-TYPE: 105P
References: <CGME20191119051537epcas5p2da5439a60d1167b8cc7e2179487996aa@epcas5p2.samsung.com>
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

Issue: In context switch, stack of next task (kernel thread)
is not mapped in previous task PGD.

Issue faced while porting VMAP stack on ARM.
currently forcible mapping is done in case of switching mm, but if
next task is kernel thread then it can cause issue.

Thus Map stack of next task in prev if next task is kernel thread,
as kernel thread will use mm of prev task.

"Since we don't have reproducible setup for x86, 
changes verified on ARM. So not sure about arch specific handling
for X86"

Signed-off-by: Vaneet Narang <v.narang@samsung.com>
Signed-off-by: Maninder Singh <maninder1.s@samsung.com>
---
 arch/x86/mm/tlb.c | 23 +++++++++++++++++++----
 1 file changed, 19 insertions(+), 4 deletions(-)

diff --git a/arch/x86/mm/tlb.c b/arch/x86/mm/tlb.c
index e6a9edc5baaf..28328cf8e79c 100644
--- a/arch/x86/mm/tlb.c
+++ b/arch/x86/mm/tlb.c
@@ -161,11 +161,17 @@ void switch_mm(struct mm_struct *prev, struct mm_struct *next,
 	local_irq_restore(flags);
 }
 
-static void sync_current_stack_to_mm(struct mm_struct *mm)
+static void sync_stack_to_mm(struct mm_struct *mm, struct task_struct *tsk)
 {
-	unsigned long sp = current_stack_pointer;
-	pgd_t *pgd = pgd_offset(mm, sp);
+	unsigned long sp;
+	pgd_t *pgd;
 
+	if (!tsk)
+		sp = current_stack_pointer;
+	else
+		sp = (unsigned long)tsk->stack;
+
+	pgd = pgd_offset(mm, sp);
 	if (pgtable_l5_enabled()) {
 		if (unlikely(pgd_none(*pgd))) {
 			pgd_t *pgd_ref = pgd_offset_k(sp);
@@ -383,7 +389,7 @@ void switch_mm_irqs_off(struct mm_struct *prev, struct mm_struct *next,
 			 * mapped in the new pgd, we'll double-fault.  Forcibly
 			 * map it.
 			 */
-			sync_current_stack_to_mm(next);
+			sync_stack_to_mm(next, NULL);
 		}
 
 		/*
@@ -460,6 +466,15 @@ void switch_mm_irqs_off(struct mm_struct *prev, struct mm_struct *next,
  */
 void enter_lazy_tlb(struct mm_struct *mm, struct task_struct *tsk)
 {
+	if (IS_ENABLED(CONFIG_VMAP_STACK)) {
+		/*
+		 * If tsk stack is in vmalloc space and isn't
+		 * mapped in the new pgd, we'll double-fault.  Forcibly
+		 * map it.
+		 */
+		sync_stack_to_mm(mm, tsk);
+	}
+
 	if (this_cpu_read(cpu_tlbstate.loaded_mm) == &init_mm)
 		return;
 
-- 
2.17.1