[S390] pfault: cpu hotplug vs missing completion interrupts

On cpu hot remove a PFAULT CANCEL command is sent to the hypervisor
which in turn will cancel all outstanding pfault requests that have
been issued on that cpu (the same happens with a SIGP cpu reset).

The result is that we end up with uninterruptible processes where
the interrupt that would wake up these processes never arrives.

In order to solve this all processes which wait for a pfault
completion interrupt get woken up after a cpu hot remove. The worst
case that could happen is that they fault again and in turn need to
wait again.

Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>

arch/s390/include/asm/lowcore.h

@@ -124,7 +124,7 @@ struct _lowcore {
 	/* Address space pointer. */
 	__u32	kernel_asce;			/* 0x02ac */
 	__u32	user_asce;			/* 0x02b0 */
-	__u8	pad_0x02b4[0x02b8-0x02b4];	/* 0x02b4 */
+	__u32	current_pid;			/* 0x02b4 */
 
 	/* SMP info area */
 	__u32	cpu_nr;				/* 0x02b8 */
@@ -255,7 +255,7 @@ struct _lowcore {
 	/* Address space pointer. */
 	__u64	kernel_asce;			/* 0x0310 */
 	__u64	user_asce;			/* 0x0318 */
-	__u8	pad_0x0320[0x0328-0x0320];	/* 0x0320 */
+	__u64	current_pid;			/* 0x0320 */
 
 	/* SMP info area */
 	__u32	cpu_nr;				/* 0x0328 */

arch/s390/include/asm/processor.h

@@ -84,6 +84,7 @@ struct thread_struct {
 	struct per_event per_event;	/* Cause of the last PER trap */
         /* pfault_wait is used to block the process on a pfault event */
 	unsigned long pfault_wait;
+	struct list_head list;
 };
 
 typedef struct thread_struct thread_struct;

arch/s390/kernel/asm-offsets.c

@@ -124,6 +124,7 @@ int main(void)
 	DEFINE(__LC_LAST_UPDATE_TIMER, offsetof(struct _lowcore, last_update_timer));
 	DEFINE(__LC_LAST_UPDATE_CLOCK, offsetof(struct _lowcore, last_update_clock));
 	DEFINE(__LC_CURRENT, offsetof(struct _lowcore, current_task));
+	DEFINE(__LC_CURRENT_PID, offsetof(struct _lowcore, current_pid));
 	DEFINE(__LC_THREAD_INFO, offsetof(struct _lowcore, thread_info));
 	DEFINE(__LC_KERNEL_STACK, offsetof(struct _lowcore, kernel_stack));
 	DEFINE(__LC_ASYNC_STACK, offsetof(struct _lowcore, async_stack));

arch/s390/kernel/entry.S

@@ -212,6 +212,7 @@ __switch_to:
 	lctl	%c4,%c4,__TASK_pid(%r3)		# load pid to control reg. 4
 	lm	%r6,%r15,__SF_GPRS(%r15)	# load gprs of next task
 	st	%r3,__LC_CURRENT		# store task struct of next
+	mvc	__LC_CURRENT_PID(4,%r0),__TASK_pid(%r3)	# store pid of next
 	st	%r5,__LC_THREAD_INFO		# store thread info of next
 	ahi	%r5,STACK_SIZE			# end of kernel stack of next
 	st	%r5,__LC_KERNEL_STACK		# store end of kernel stack

arch/s390/kernel/entry64.S

@@ -220,6 +220,7 @@ __switch_to:
 	lctl	%c4,%c4,__TASK_pid(%r3)		# load pid to control reg. 4
 	lmg	%r6,%r15,__SF_GPRS(%r15)	# load gprs of next task
 	stg	%r3,__LC_CURRENT		# store task struct of next
+	mvc	__LC_CURRENT_PID+4(4,%r0),__TASK_pid(%r3) # store pid of next
 	stg	%r5,__LC_THREAD_INFO		# store thread info of next
 	aghi	%r5,STACK_SIZE			# end of kernel stack of next
 	stg	%r5,__LC_KERNEL_STACK		# store end of kernel stack

arch/s390/mm/fault.c

@@ -466,7 +466,7 @@ typedef struct {
 int pfault_init(void)
 {
 	pfault_refbk_t refbk =
-		{ 0x258, 0, 5, 2, __LC_CURRENT, 1ULL << 48, 1ULL << 48,
+		{ 0x258, 0, 5, 2, __LC_CURRENT_PID, 1ULL << 48, 1ULL << 48,
 		  __PF_RES_FIELD };
         int rc;
 
@@ -498,11 +498,15 @@ void pfault_fini(void)
 		: : "a" (&refbk), "m" (refbk) : "cc");
 }
 
+static DEFINE_SPINLOCK(pfault_lock);
+static LIST_HEAD(pfault_list);
+
 static void pfault_interrupt(unsigned int ext_int_code,
 			     unsigned int param32, unsigned long param64)
 {
 	struct task_struct *tsk;
 	__u16 subcode;
+	pid_t pid;
 
 	/*
 	 * Get the external interruption subcode & pfault
@@ -514,44 +518,79 @@ static void pfault_interrupt(unsigned int ext_int_code,
 	if ((subcode & 0xff00) != __SUBCODE_MASK)
 		return;
 	kstat_cpu(smp_processor_id()).irqs[EXTINT_PFL]++;
-
-	/*
-	 * Get the token (= address of the task structure of the affected task).
-	 */
-#ifdef CONFIG_64BIT
-	tsk = (struct task_struct *) param64;
-#else
-	tsk = (struct task_struct *) param32;
-#endif
-
+	if (subcode & 0x0080) {
+		/* Get the token (= pid of the affected task). */
+		pid = sizeof(void *) == 4 ? param32 : param64;
+		rcu_read_lock();
+		tsk = find_task_by_pid_ns(pid, &init_pid_ns);
+		if (tsk)
+			get_task_struct(tsk);
+		rcu_read_unlock();
+		if (!tsk)
+			return;
+	} else {
+		tsk = current;
+	}
+	spin_lock(&pfault_lock);
 	if (subcode & 0x0080) {
 		/* signal bit is set -> a page has been swapped in by VM */
-		if (xchg(&tsk->thread.pfault_wait, -1) != 0) {
+		if (tsk->thread.pfault_wait == 1) {
 			/* Initial interrupt was faster than the completion
 			 * interrupt. pfault_wait is valid. Set pfault_wait
 			 * back to zero and wake up the process. This can
 			 * safely be done because the task is still sleeping
 			 * and can't produce new pfaults. */
 			tsk->thread.pfault_wait = 0;
+			list_del(&tsk->thread.list);
 			wake_up_process(tsk);
-			put_task_struct(tsk);
+		} else {
+			/* Completion interrupt was faster than initial
+			 * interrupt. Set pfault_wait to -1 so the initial
+			 * interrupt doesn't put the task to sleep. */
+			tsk->thread.pfault_wait = -1;
 		}
+		put_task_struct(tsk);
 	} else {
 		/* signal bit not set -> a real page is missing. */
-		get_task_struct(tsk);
-		set_task_state(tsk, TASK_UNINTERRUPTIBLE);
-		if (xchg(&tsk->thread.pfault_wait, 1) != 0) {
+		if (tsk->thread.pfault_wait == -1) {
 			/* Completion interrupt was faster than the initial
-			 * interrupt (swapped in a -1 for pfault_wait). Set
-			 * pfault_wait back to zero and exit. This can be
-			 * done safely because tsk is running in kernel 
-			 * mode and can't produce new pfaults. */
+			 * interrupt (pfault_wait == -1). Set pfault_wait
+			 * back to zero and exit. */
 			tsk->thread.pfault_wait = 0;
-			set_task_state(tsk, TASK_RUNNING);
-			put_task_struct(tsk);
-		} else
+		} else {
+			/* Initial interrupt arrived before completion
+			 * interrupt. Let the task sleep. */
+			tsk->thread.pfault_wait = 1;
+			list_add(&tsk->thread.list, &pfault_list);
+			set_task_state(tsk, TASK_UNINTERRUPTIBLE);
 			set_tsk_need_resched(tsk);
+		}
 	}
+	spin_unlock(&pfault_lock);
+}
+
+static int __cpuinit pfault_cpu_notify(struct notifier_block *self,
+				       unsigned long action, void *hcpu)
+{
+	struct thread_struct *thread, *next;
+	struct task_struct *tsk;
+
+	switch (action) {
+	case CPU_DEAD:
+	case CPU_DEAD_FROZEN:
+		spin_lock_irq(&pfault_lock);
+		list_for_each_entry_safe(thread, next, &pfault_list, list) {
+			thread->pfault_wait = 0;
+			list_del(&thread->list);
+			tsk = container_of(thread, struct task_struct, thread);
+			wake_up_process(tsk);
+		}
+		spin_unlock_irq(&pfault_lock);
+		break;
+	default:
+		break;
+	}
+	return NOTIFY_OK;
 }
 
 static int __init pfault_irq_init(void)
@@ -568,8 +607,10 @@ static int __init pfault_irq_init(void)
 		pfault_disable = 1;
 		return rc;
 	}
-	if (pfault_init() == 0)
+	if (pfault_init() == 0) {
+		hotcpu_notifier(pfault_cpu_notify, 0);
 		return 0;
+	}
 
 	/* Tough luck, no pfault. */
 	pfault_disable = 1;