Skip to contentSkip to conversation

sparc32: destroy_context() and switch_mm() needs to disable interrupts.

PATCH In review Andreas Larsson 2 participants 2 messages Dec 18, 2014 SPARC + UltraSPARC (sparc/sparc64)

lore
Share & tools
lore

Andreas Larsson

author maintainer
#0 lore raw

PATCH 1/1 sparc32: destroy_context() and switch_mm() needs to disable interrupts.

Load balancing can be triggered in the critical sections protected by srmmu_context_spinlock in destroy_context() and switch_mm() and can hang the cpu waiting for the rq lock of another cpu that in turn has called switch_mm hangning on srmmu_context_spinlock leading to deadlock.

So, disable interrupt while taking srmmu_context_spinlock in destroy_context() and switch_mm() so we don't deadlock.

See also commit 77b838fa1ef0 ("[SPARC64]: destroy_context() needs to disable interrupts.")

---
 arch/sparc/mm/srmmu.c |   11 +++++++----
 1 file changed, 7 insertions(+), 4 deletions(-)
arch/sparc/mm/srmmu.c +7 -4
index be65f03..5cbc96d 100644
@@ -460,10 +460,12 @@ static void __init sparc_context_init(int numctx)
 void switch_mm(struct mm_struct *old_mm, struct mm_struct *mm,
 	       struct task_struct *tsk)
 {
+	unsigned long flags;
+
 	if (mm->context = NO_CONTEXT) {
-		spin_lock(&srmmu_context_spinlock);
+		spin_lock_irqsave(&srmmu_context_spinlock, flags);
 		alloc_context(old_mm, mm);
-		spin_unlock(&srmmu_context_spinlock);
+		spin_unlock_irqrestore(&srmmu_context_spinlock, flags);
 		srmmu_ctxd_set(&srmmu_context_table[mm->context], mm->pgd);
 	}
 
@@ -986,14 +988,15 @@ int init_new_context(struct task_struct *tsk, struct mm_struct *mm)
 
 void destroy_context(struct mm_struct *mm)
 {
+	unsigned long flags;
 
 	if (mm->context != NO_CONTEXT) {
 		flush_cache_mm(mm);
 		srmmu_ctxd_set(&srmmu_context_table[mm->context], srmmu_swapper_pg_dir);
 		flush_tlb_mm(mm);
-		spin_lock(&srmmu_context_spinlock);
+		spin_lock_irqsave(&srmmu_context_spinlock, flags);
 		free_context(mm->context);
-		spin_unlock(&srmmu_context_spinlock);
+		spin_unlock_irqrestore(&srmmu_context_spinlock, flags);
 		mm->context = NO_CONTEXT;
 	}
 }
Signature · 1 line
1.7.10.4
maintainer
#1 lore raw

From: Andreas Larsson <[email protected]>
Date: Thu, 18 Dec 2014 13:23:23 +0100

Quoted 5 lines

Load balancing can be triggered in the critical sections protected by srmmu_context_spinlock in destroy_context() and switch_mm() and can hang the cpu waiting for the rq lock of another cpu that in turn has called switch_mm hangning on srmmu_context_spinlock leading to deadlock.

Quoted 7 lines

So, disable interrupt while taking srmmu_context_spinlock in destroy_context() and switch_mm() so we don't deadlock.

See also commit 77b838fa1ef0 ("[SPARC64]: destroy_context() needs to disable interrupts.")

Looks good, applied, thanks Andreas.