Skip to contentSkip to conversation

PS3: Clear SPU Class 0 interrupts in handler

PATCH In review Geoff Levand 2 participants 2 messages Sep 4, 2007

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/2 PS3: Clear SPU Class 0 interrupts in handler

From: Masato Noguchi <[email protected]>

Fix a bug that causes the PS3 to hang on the SPU Class 0 interrupt.

The Cell BE Architecture spec states that the SPU MFC Class 0 interrupt is delivered as a pulse. The current spu interrupt handler assumes this behavior and does not clear the interrupt status.

The PS3 hypervisor visualizes all SPU interrupts as level, and on return from the interrupt handler the hypervisor will deliver a new virtual interrupt for any unmasked interrupts which for which the status has not been cleared. This fix clears the interrupt status in the interrupt handler.

---

Arnd, please consider for 2.6.24.
Tested on QS20, and seems OK.

 arch/powerpc/platforms/cell/spu_base.c |   23 +++++++++++++++--------
 include/asm-powerpc/spu.h              |    2 +-
 2 files changed, 16 insertions(+), 9 deletions(-)
arch/powerpc/platforms/cell/spu_base.c +15 -8
@@ -236,27 +236,34 @@ static irqreturn_t
 spu_irq_class_0(int irq, void *data)
 {
 	struct spu *spu;
+	unsigned long stat, mask;
 
 	spu = data;
-	spu->class_0_pending = 1;
+
+	mask = spu_int_mask_get(spu, 0);
+	stat = spu_int_stat_get(spu, 0);
+	stat &= mask;
+
+	spin_lock(&spu->register_lock);
+	spu->class_0_pending |= stat;
+	spin_unlock(&spu->register_lock);
+
 	spu->stop_callback(spu);
 
+	spu_int_stat_clear(spu, 0, stat);
+
 	return IRQ_HANDLED;
 }
 
 int
 spu_irq_class_0_bottom(struct spu *spu)
 {
-	unsigned long stat, mask;
 	unsigned long flags;
-
-	spu->class_0_pending = 0;
+	unsigned long stat;
 
 	spin_lock_irqsave(&spu->register_lock, flags);
-	mask = spu_int_mask_get(spu, 0);
-	stat = spu_int_stat_get(spu, 0);
-
-	stat &= mask;
+	stat = spu->class_0_pending;
+	spu->class_0_pending = 0;
 
 	if (stat & 1) /* invalid DMA alignment */
 		__spu_trap_dma_align(spu);
include/asm-powerpc/spu.h +1 -1
@@ -130,6 +130,7 @@ struct spu {
 	u64 flags;
 	u64 dar;
 	u64 dsisr;
+	u64 class_0_pending;
 	size_t ls_size;
 	unsigned int slb_replace;
 	struct mm_struct *mm;
@@ -138,7 +139,6 @@ struct spu {
 	unsigned long long timestamp;
 	pid_t pid;
 	pid_t tgid;
-	int class_0_pending;
 	spinlock_t register_lock;
 
 	void (* wbox_callback)(struct spu *spu);
Acked-by on v1 1/2
#1 lore raw
Quoted 6 lines · Geoff Levand

From: Masato Noguchi <[email protected]>
=20
Fix a bug that causes the PS3 to hang on the SPU Class 0 interrupt. =20
The Cell BE Architecture spec states that the SPU MFC Class 0 interrupt is delivered as a pulse. =C2=A0The current spu interrupt handler assumes =

this

Quoted 7 lines

behavior and does not clear the interrupt status.
=20
The PS3 hypervisor visualizes all SPU interrupts as level, and on return from the interrupt handler the hypervisor will deliver a new virtual interrupt for any unmasked interrupts which for which the status has not been cleared. =C2=A0This fix clears the interrupt status in the interrupt handler.

Right, we have this one in the sdk3.0/2.6.22-arnd2 kernel since I pulled in all of the ps3-linux tree, so it was already part of our testing, and the explanation makes sense, so we should certainly have it upstream.

Jeremy, please add this to your patch queue.

Quoted 2 lines