From 5711fe900dfef8d9afdbbb6d0f9c9720919d1d66 Mon Sep 17 00:00:00 2001
From: Jeremy Kerr <jk@ozlabs.org>
Date: Fri, 4 Apr 2008 17:55:28 +1100
Subject: [POWERPC] cell: Fix lost interrupts due to fasteoi handler

We may currently lose interrupts during SPE context switch, as we alter
the INT_Route register. Because the IIC uses a per-thread priority
status, changing the interrupt routing to a different thread means that
the IRQ is no longer masked by the priority status, so we end up with
two fasteoi IRQ handlers executing for the one irq_desc. The fasteoi
handler doesn't handle multiple IRQs, so drops the second one.

Fix this by using our own flow handler. This is based on
handle_edge_irq, but issues an eoi after IRQs are handled, and doesn't
do any mask/unmasking.

Signed-off-by: Jeremy Kerr <jk@ozlabs.org>
---
 arch/powerpc/platforms/cell/interrupt.c | 53 +++++++++++++++++++++++++++++++--
 1 file changed, 51 insertions(+), 2 deletions(-)

(limited to 'arch/powerpc/platforms')

diff --git a/arch/powerpc/platforms/cell/interrupt.c b/arch/powerpc/platforms/cell/interrupt.c
index 04f74f9f9ab..5bf7df14602 100644
--- a/arch/powerpc/platforms/cell/interrupt.c
+++ b/arch/powerpc/platforms/cell/interrupt.c
@@ -35,6 +35,7 @@
 #include <linux/percpu.h>
 #include <linux/types.h>
 #include <linux/ioport.h>
+#include <linux/kernel_stat.h>
 
 #include <asm/io.h>
 #include <asm/pgtable.h>
@@ -231,6 +232,54 @@ static int iic_host_match(struct irq_host *h, struct device_node *node)
 				    "IBM,CBEA-Internal-Interrupt-Controller");
 }
 
+extern int noirqdebug;
+
+static void handle_iic_irq(unsigned int irq, struct irq_desc *desc)
+{
+	const unsigned int cpu = smp_processor_id();
+
+	spin_lock(&desc->lock);
+
+	desc->status &= ~(IRQ_REPLAY | IRQ_WAITING);
+
+	/*
+	 * If we're currently running this IRQ, or its disabled,
+	 * we shouldn't process the IRQ. Mark it pending, handle
+	 * the necessary masking and go out
+	 */
+	if (unlikely((desc->status & (IRQ_INPROGRESS | IRQ_DISABLED)) ||
+		    !desc->action)) {
+		desc->status |= IRQ_PENDING;
+		goto out_eoi;
+	}
+
+	kstat_cpu(cpu).irqs[irq]++;
+
+	/* Mark the IRQ currently in progress.*/
+	desc->status |= IRQ_INPROGRESS;
+
+	do {
+		struct irqaction *action = desc->action;
+		irqreturn_t action_ret;
+
+		if (unlikely(!action))
+			goto out_eoi;
+
+		desc->status &= ~IRQ_PENDING;
+		spin_unlock(&desc->lock);
+		action_ret = handle_IRQ_event(irq, action);
+		if (!noirqdebug)
+			note_interrupt(irq, desc, action_ret);
+		spin_lock(&desc->lock);
+
+	} while ((desc->status & (IRQ_PENDING | IRQ_DISABLED)) == IRQ_PENDING);
+
+	desc->status &= ~IRQ_INPROGRESS;
+out_eoi:
+	desc->chip->eoi(irq);
+	spin_unlock(&desc->lock);
+}
+
 static int iic_host_map(struct irq_host *h, unsigned int virq,
 			irq_hw_number_t hw)
 {
@@ -240,10 +289,10 @@ static int iic_host_map(struct irq_host *h, unsigned int virq,
 		break;
 	case IIC_IRQ_TYPE_IOEXC:
 		set_irq_chip_and_handler(virq, &iic_ioexc_chip,
-					 handle_fasteoi_irq);
+					 handle_iic_irq);
 		break;
 	default:
-		set_irq_chip_and_handler(virq, &iic_chip, handle_fasteoi_irq);
+		set_irq_chip_and_handler(virq, &iic_chip, handle_iic_irq);
 	}
 	return 0;
 }
-- 
cgit v1.2.3


From 943906ba4bebf629d5cd770e48b8ec0ddc433869 Mon Sep 17 00:00:00 2001
From: Jeremy Kerr <jk@ozlabs.org>
Date: Mon, 7 Apr 2008 17:42:36 +1000
Subject: [POWERPC] spufs: don't touch suspend bits when purging DMA queue

When we issue a MFC purge request, we may inadvertantly clear the
suspended status.

This change adds the MFC_CNTL_SUSPEND_MASK when we issue a purge
request, so that the suspend bit is masked out.

Signed-off-by: Jeremy Kerr <jk@ozlabs.org>
---
 arch/powerpc/platforms/cell/spufs/switch.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

(limited to 'arch/powerpc/platforms')

diff --git a/arch/powerpc/platforms/cell/spufs/switch.c b/arch/powerpc/platforms/cell/spufs/switch.c
index d2a1249d36d..65b423b6da4 100644
--- a/arch/powerpc/platforms/cell/spufs/switch.c
+++ b/arch/powerpc/platforms/cell/spufs/switch.c
@@ -462,7 +462,9 @@ static inline void purge_mfc_queue(struct spu_state *csa, struct spu *spu)
 	 * Restore, Step 14.
 	 *     Write MFC_CNTL[Pc]=1 (purge queue).
 	 */
-	out_be64(&priv2->mfc_control_RW, MFC_CNTL_PURGE_DMA_REQUEST);
+	out_be64(&priv2->mfc_control_RW,
+			MFC_CNTL_PURGE_DMA_REQUEST |
+			MFC_CNTL_SUSPEND_MASK);
 	eieio();
 }
 
-- 
cgit v1.2.3


From 1ca4264ee17745779c341966c5e61ac69bfd17a6 Mon Sep 17 00:00:00 2001
From: Jeremy Kerr <jk@ozlabs.org>
Date: Wed, 9 Apr 2008 15:24:22 +1000
Subject: [POWERPC] spufs: fix save of mfc_cntl register

Currently, we can introduce invalid entries into the MFC queues:

1) context starts a DMA

2) context gets scheduled out during a DMA
  - kernel saves MFC queue to CSA
  - kernel saves 0x0 in csa->mfc_control_RW

3) context gets scheduled in
  - csa->mfc_control[Q] ('queues empty') isn't set, so DMA queues are
    restored from the CSA

4) context's DMA is completed

5) context gets scheduled out again, no DMA occuring this time
  - kernel sees that MFC_CNTL[Q] ('queues empty') is set, so doesn't
    touch saved queue data in CSA
  - kernel saves 0x0 in csa->mfc_control_RW

6) context gets scheduled in
  - csa->mfc_control[Q] ('queues empty') isn't set (we saved is as 0!),
    so DMA queues are restored from the CSA

In this last restore, we've restored the queue status from step 2,
which are now invalid.

This change makes save_mfc_cntl() closer to the save/restore sequence,
as specified in the CBE handbook.

With changes from Luke Browning.

Signed-off-by: Jeremy Kerr <jk@ozlabs.org>
---
 arch/powerpc/platforms/cell/spufs/switch.c | 13 +++++++------
 1 file changed, 7 insertions(+), 6 deletions(-)

(limited to 'arch/powerpc/platforms')

diff --git a/arch/powerpc/platforms/cell/spufs/switch.c b/arch/powerpc/platforms/cell/spufs/switch.c
index 65b423b6da4..4131784216b 100644
--- a/arch/powerpc/platforms/cell/spufs/switch.c
+++ b/arch/powerpc/platforms/cell/spufs/switch.c
@@ -186,20 +186,21 @@ static inline void save_mfc_cntl(struct spu_state *csa, struct spu *spu)
 				 MFC_CNTL_SUSPEND_COMPLETE);
 		/* fall through */
 	case MFC_CNTL_SUSPEND_COMPLETE:
-		if (csa) {
+		if (csa)
 			csa->priv2.mfc_control_RW =
-				MFC_CNTL_SUSPEND_MASK |
+				in_be64(&priv2->mfc_control_RW) |
 				MFC_CNTL_SUSPEND_DMA_QUEUE;
-		}
 		break;
 	case MFC_CNTL_NORMAL_DMA_QUEUE_OPERATION:
 		out_be64(&priv2->mfc_control_RW, MFC_CNTL_SUSPEND_DMA_QUEUE);
 		POLL_WHILE_FALSE((in_be64(&priv2->mfc_control_RW) &
 				  MFC_CNTL_SUSPEND_DMA_STATUS_MASK) ==
 				 MFC_CNTL_SUSPEND_COMPLETE);
-		if (csa) {
-			csa->priv2.mfc_control_RW = 0;
-		}
+		if (csa)
+			csa->priv2.mfc_control_RW =
+				in_be64(&priv2->mfc_control_RW) &
+				~MFC_CNTL_SUSPEND_DMA_QUEUE &
+				~MFC_CNTL_SUSPEND_MASK;
 		break;
 	}
 }
-- 
cgit v1.2.3


From 55d7cd74d44e7e0597f95db25c12f3e6c5e7916f Mon Sep 17 00:00:00 2001
From: Jeremy Kerr <jk@ozlabs.org>
Date: Thu, 24 Apr 2008 18:15:40 +1000
Subject: [POWERPC] spufs: fix post-stopped update of MFC_CNTL register

We currently have two issues with the MFC save code:

 * save_mfc_decr doesn't handle a transition of 1 -> 0 of the Ds bit
 * The Q bit may be stale in the CSA

This change fixes the first issue by clearing the relevant bits from
the MFC_CNTL value in the CSA before or-ing in the updated status.
Also, we add the Q bit to the updated status.

Signed-off-by: Jeremy Kerr <jk@ozlabs.org>
---
 arch/powerpc/platforms/cell/spufs/switch.c | 13 +++++++++----
 1 file changed, 9 insertions(+), 4 deletions(-)

(limited to 'arch/powerpc/platforms')

diff --git a/arch/powerpc/platforms/cell/spufs/switch.c b/arch/powerpc/platforms/cell/spufs/switch.c
index 4131784216b..53361d0c070 100644
--- a/arch/powerpc/platforms/cell/spufs/switch.c
+++ b/arch/powerpc/platforms/cell/spufs/switch.c
@@ -250,16 +250,21 @@ static inline void save_spu_status(struct spu_state *csa, struct spu *spu)
 	}
 }
 
-static inline void save_mfc_decr(struct spu_state *csa, struct spu *spu)
+static inline void save_mfc_stopped_status(struct spu_state *csa,
+		struct spu *spu)
 {
 	struct spu_priv2 __iomem *priv2 = spu->priv2;
+	const u64 mask = MFC_CNTL_DECREMENTER_RUNNING |
+			MFC_CNTL_DMA_QUEUES_EMPTY;
 
 	/* Save, Step 12:
 	 *     Read MFC_CNTL[Ds].  Update saved copy of
 	 *     CSA.MFC_CNTL[Ds].
+	 *
+	 * update: do the same with MFC_CNTL[Q].
 	 */
-	csa->priv2.mfc_control_RW |=
-		in_be64(&priv2->mfc_control_RW) & MFC_CNTL_DECREMENTER_RUNNING;
+	csa->priv2.mfc_control_RW &= ~mask;
+	csa->priv2.mfc_control_RW |= in_be64(&priv2->mfc_control_RW) & mask;
 }
 
 static inline void halt_mfc_decr(struct spu_state *csa, struct spu *spu)
@@ -1791,7 +1796,7 @@ static int quiece_spu(struct spu_state *prev, struct spu *spu)
 	save_spu_runcntl(prev, spu);	        /* Step 9. */
 	save_mfc_sr1(prev, spu);	        /* Step 10. */
 	save_spu_status(prev, spu);	        /* Step 11. */
-	save_mfc_decr(prev, spu);	        /* Step 12. */
+	save_mfc_stopped_status(prev, spu);     /* Step 12. */
 	halt_mfc_decr(prev, spu);	        /* Step 13. */
 	save_timebase(prev, spu);		/* Step 14. */
 	remove_other_spu_access(prev, spu);	/* Step 15. */
-- 
cgit v1.2.3


From c0bace5c7029ef7cbb6e6ffaa0469a3c042e5029 Mon Sep 17 00:00:00 2001
From: Jeremy Kerr <jk@ozlabs.org>
Date: Wed, 23 Apr 2008 14:24:27 +1000
Subject: [POWERPC] spufs: update master runcntl with context lock held

Currently, we update the SPU master run control bit (ie,
spu_enable_spu) in spufs_run_spu before we grab the context mutex. This
can result in races with other processes accessing this context's
resources.

This change moves the spu_enable_spu to after we have acquired the
context lock.

Signed-off-by: Jeremy Kerr <jk@ozlabs.org>
---
 arch/powerpc/platforms/cell/spufs/run.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

(limited to 'arch/powerpc/platforms')

diff --git a/arch/powerpc/platforms/cell/spufs/run.c b/arch/powerpc/platforms/cell/spufs/run.c
index a9c35b7b719..0634fecfe54 100644
--- a/arch/powerpc/platforms/cell/spufs/run.c
+++ b/arch/powerpc/platforms/cell/spufs/run.c
@@ -343,13 +343,14 @@ long spufs_run_spu(struct spu_context *ctx, u32 *npc, u32 *event)
 	if (mutex_lock_interruptible(&ctx->run_mutex))
 		return -ERESTARTSYS;
 
-	spu_enable_spu(ctx);
 	ctx->event_return = 0;
 
 	ret = spu_acquire(ctx);
 	if (ret)
 		goto out_unlock;
 
+	spu_enable_spu(ctx);
+
 	spu_update_sched_info(ctx);
 
 	ret = spu_run_init(ctx, npc);
-- 
cgit v1.2.3


From d29694f0681e188cf75477f20904041744e78ef7 Mon Sep 17 00:00:00 2001
From: Jeremy Kerr <jk@ozlabs.org>
Date: Wed, 23 Apr 2008 16:02:10 +1000
Subject: [POWERPC] spufs: don't acquire state_mutex interruptible while
 performing callback

There's currently no way to tell if spu_process_callback has
returned with the state mutex held, as -EINTR may be returned
by either the syscall or the spu_acquire fail case.

Instead, just do a non-interruptible mutex_lock here.

Signed-off-by: Jeremy Kerr <jk@ozlabs.org>
---
 arch/powerpc/platforms/cell/spufs/run.c | 6 ++----
 1 file changed, 2 insertions(+), 4 deletions(-)

(limited to 'arch/powerpc/platforms')

diff --git a/arch/powerpc/platforms/cell/spufs/run.c b/arch/powerpc/platforms/cell/spufs/run.c
index 0634fecfe54..e764a43b544 100644
--- a/arch/powerpc/platforms/cell/spufs/run.c
+++ b/arch/powerpc/platforms/cell/spufs/run.c
@@ -294,7 +294,7 @@ static int spu_process_callback(struct spu_context *ctx)
 	u32 ls_pointer, npc;
 	void __iomem *ls;
 	long spu_ret;
-	int ret, ret2;
+	int ret;
 
 	/* get syscall block from local store */
 	npc = ctx->ops->npc_read(ctx) & ~3;
@@ -316,11 +316,9 @@ static int spu_process_callback(struct spu_context *ctx)
 		if (spu_ret <= -ERESTARTSYS) {
 			ret = spu_handle_restartsys(ctx, &spu_ret, &npc);
 		}
-		ret2 = spu_acquire(ctx);
+		mutex_lock(&ctx->state_mutex);
 		if (ret == -ERESTARTSYS)
 			return ret;
-		if (ret2)
-			return -EINTR;
 	}
 
 	/* need to re-get the ls, as it may have changed when we released the
-- 
cgit v1.2.3


From 093c16bf9b107433643cbf0843ca7808df90823b Mon Sep 17 00:00:00 2001
From: Luke Browning <lukebr@linux.vnet.ibm.com>
Date: Fri, 25 Apr 2008 14:00:00 +1000
Subject: [POWERPC] spufs: set SPU_CONTEXT_SWITCH_PENDING before synchronising
 SPU irqs

synchronize_irq() provides the serialization for
SPU_CONTEXT_SWITCH_PENDING which is read with a simple load. This
routine guarantees that the relevant interrupt handlers are not running,
so that the next time they do run they will see the update
memory value.

This must be done correctly so that exception handling code does not
restart the mfc in the middle of a context switch while we are trying
to atomically stop it and save state.

Signed-off-by: Luke Browning <lukebr@linux.vnet.ibm.com>
Signed-off-by: Jeremy Kerr <jk@ozlabs.org>
---
 arch/powerpc/platforms/cell/spufs/switch.c | 10 ++++++++--
 1 file changed, 8 insertions(+), 2 deletions(-)

(limited to 'arch/powerpc/platforms')

diff --git a/arch/powerpc/platforms/cell/spufs/switch.c b/arch/powerpc/platforms/cell/spufs/switch.c
index 53361d0c070..9748808b0f1 100644
--- a/arch/powerpc/platforms/cell/spufs/switch.c
+++ b/arch/powerpc/platforms/cell/spufs/switch.c
@@ -132,6 +132,13 @@ static inline void disable_interrupts(struct spu_state *csa, struct spu *spu)
 	spu_int_mask_set(spu, 2, 0ul);
 	eieio();
 	spin_unlock_irq(&spu->register_lock);
+
+	/*
+	 * This flag needs to be set before calling synchronize_irq so
+	 * that the update will be visible to the relevant handlers
+	 * via a simple load.
+	 */
+	set_bit(SPU_CONTEXT_SWITCH_PENDING, &spu->flags);
 	synchronize_irq(spu->irqs[0]);
 	synchronize_irq(spu->irqs[1]);
 	synchronize_irq(spu->irqs[2]);
@@ -166,9 +173,8 @@ static inline void set_switch_pending(struct spu_state *csa, struct spu *spu)
 	/* Save, Step 7:
 	 * Restore, Step 5:
 	 *     Set a software context switch pending flag.
+	 *     Done above in Step 3 - disable_interrupts().
 	 */
-	set_bit(SPU_CONTEXT_SWITCH_PENDING, &spu->flags);
-	mb();
 }
 
 static inline void save_mfc_cntl(struct spu_state *csa, struct spu *spu)
-- 
cgit v1.2.3


From 7a2142002f29a7b398c49da9bdec712dc57087c7 Mon Sep 17 00:00:00 2001
From: Luke Browning <lukebr@linux.vnet.ibm.com>
Date: Mon, 28 Apr 2008 14:32:34 +1000
Subject: [POWERPC] spufs: try to route SPU interrupts to local node

Currently, we re-route SPU interrupts to the current cpu, which may be
on a remote node. In the case of time slicing, all spu interrupts will
end up routed to the same cpu, where the spusched_tick occurs.

This change routes mfc interrupts to the cpu where the controlling
thread last ran, provided that cpu is on the same node as the spu
(otherwise don't reroute interrupts).

This should improve performance and provide a more predictable
environment for processing spu exceptions. In the past we have seen
concurrent delivery of spu exceptions to two cpus. This eliminates that
concern.

Signed-off-by: Luke Browning <lukebr@linux.vnet.ibm.com>
Signed-off-by: Jeremy Kerr <jk@ozlabs.org>
---
 arch/powerpc/platforms/cell/spu_priv1_mmio.c | 16 ++++++++++++++--
 arch/powerpc/platforms/cell/spufs/sched.c    |  4 +++-
 arch/powerpc/platforms/cell/spufs/spufs.h    |  1 +
 arch/powerpc/platforms/cell/spufs/switch.c   |  8 ++++++++
 4 files changed, 26 insertions(+), 3 deletions(-)

(limited to 'arch/powerpc/platforms')

diff --git a/arch/powerpc/platforms/cell/spu_priv1_mmio.c b/arch/powerpc/platforms/cell/spu_priv1_mmio.c
index 67fa7247b80..906a0a2a9fe 100644
--- a/arch/powerpc/platforms/cell/spu_priv1_mmio.c
+++ b/arch/powerpc/platforms/cell/spu_priv1_mmio.c
@@ -28,6 +28,7 @@
 #include <linux/io.h>
 #include <linux/mutex.h>
 #include <linux/device.h>
+#include <linux/sched.h>
 
 #include <asm/spu.h>
 #include <asm/spu_priv1.h>
@@ -75,8 +76,19 @@ static u64 int_stat_get(struct spu *spu, int class)
 
 static void cpu_affinity_set(struct spu *spu, int cpu)
 {
-	u64 target = iic_get_target_id(cpu);
-	u64 route = target << 48 | target << 32 | target << 16;
+	u64 target;
+	u64 route;
+
+	if (nr_cpus_node(spu->node)) {
+		cpumask_t spumask = node_to_cpumask(spu->node);
+		cpumask_t cpumask = node_to_cpumask(cpu_to_node(cpu));
+
+		if (!cpus_intersects(spumask, cpumask))
+			return;
+	}
+
+	target = iic_get_target_id(cpu);
+	route = target << 48 | target << 32 | target << 16;
 	out_be64(&spu->priv1->int_route_RW, route);
 }
 
diff --git a/arch/powerpc/platforms/cell/spufs/sched.c b/arch/powerpc/platforms/cell/spufs/sched.c
index 62280c292aa..1df7d6d152c 100644
--- a/arch/powerpc/platforms/cell/spufs/sched.c
+++ b/arch/powerpc/platforms/cell/spufs/sched.c
@@ -140,6 +140,9 @@ void __spu_update_sched_info(struct spu_context *ctx)
 	 * if it is timesliced or preempted.
 	 */
 	ctx->cpus_allowed = current->cpus_allowed;
+
+	/* Save the current cpu id for spu interrupt routing. */
+	ctx->last_ran = raw_smp_processor_id();
 }
 
 void spu_update_sched_info(struct spu_context *ctx)
@@ -243,7 +246,6 @@ static void spu_bind_context(struct spu *spu, struct spu_context *ctx)
 	spu_switch_log_notify(spu, ctx, SWITCH_LOG_START, 0);
 	spu_restore(&ctx->csa, spu);
 	spu->timestamp = jiffies;
-	spu_cpu_affinity_set(spu, raw_smp_processor_id());
 	spu_switch_notify(spu, ctx);
 	ctx->state = SPU_STATE_RUNNABLE;
 
diff --git a/arch/powerpc/platforms/cell/spufs/spufs.h b/arch/powerpc/platforms/cell/spufs/spufs.h
index 7312745b754..dc3a215a6a2 100644
--- a/arch/powerpc/platforms/cell/spufs/spufs.h
+++ b/arch/powerpc/platforms/cell/spufs/spufs.h
@@ -121,6 +121,7 @@ struct spu_context {
 	cpumask_t cpus_allowed;
 	int policy;
 	int prio;
+	int last_ran;
 
 	/* statistics */
 	struct {
diff --git a/arch/powerpc/platforms/cell/spufs/switch.c b/arch/powerpc/platforms/cell/spufs/switch.c
index 9748808b0f1..47c658051fc 100644
--- a/arch/powerpc/platforms/cell/spufs/switch.c
+++ b/arch/powerpc/platforms/cell/spufs/switch.c
@@ -1704,6 +1704,13 @@ static inline void restore_mfc_sr1(struct spu_state *csa, struct spu *spu)
 	eieio();
 }
 
+static inline void set_int_route(struct spu_state *csa, struct spu *spu)
+{
+	struct spu_context *ctx = spu->ctx;
+
+	spu_cpu_affinity_set(spu, ctx->last_ran);
+}
+
 static inline void restore_other_spu_access(struct spu_state *csa,
 					    struct spu *spu)
 {
@@ -2014,6 +2021,7 @@ static void restore_csa(struct spu_state *next, struct spu *spu)
 	check_ppuint_mb_stat(next, spu);	/* Step 67. */
 	spu_invalidate_slbs(spu);		/* Modified Step 68. */
 	restore_mfc_sr1(next, spu);	        /* Step 69. */
+	set_int_route(next, spu);		/* NEW      */
 	restore_other_spu_access(next, spu);	/* Step 70. */
 	restore_spu_runcntl(next, spu);	        /* Step 71. */
 	restore_mfc_cntl(next, spu);	        /* Step 72. */
-- 
cgit v1.2.3


From f3d69e0507f84903059d456c5d19f10b2df3ac69 Mon Sep 17 00:00:00 2001
From: Luke Browning <lukebr@linux.vnet.ibm.com>
Date: Sun, 27 Apr 2008 18:41:55 +0000
Subject: [POWERPC] spufs: fix concurrent delivery of class 0 & 1 exceptions

SPU class 0 & 1 exceptions may occur in parallel, so we may end up
overwriting csa.dsisr.

This change adds dedicated fields for each class to the spu and the spu
context so that fault data is not overwritten.

Signed-off-by: Luke Browning <lukebr@linux.vnet.ibm.com>
Signed-off-by: Jeremy Kerr <jk@ozlabs.org>
---
 arch/powerpc/platforms/cell/spu_base.c    | 27 ++++++++++++++++++---------
 arch/powerpc/platforms/cell/spufs/fault.c | 17 +++++++++++------
 arch/powerpc/platforms/cell/spufs/run.c   | 29 +++++++++++++++++++----------
 arch/powerpc/platforms/cell/spufs/spufs.h |  2 +-
 4 files changed, 49 insertions(+), 26 deletions(-)

(limited to 'arch/powerpc/platforms')

diff --git a/arch/powerpc/platforms/cell/spu_base.c b/arch/powerpc/platforms/cell/spu_base.c
index 6bab44b7716..b9ae675640d 100644
--- a/arch/powerpc/platforms/cell/spu_base.c
+++ b/arch/powerpc/platforms/cell/spu_base.c
@@ -226,11 +226,13 @@ static int __spu_trap_data_map(struct spu *spu, unsigned long ea, u64 dsisr)
 		return 0;
 	}
 
-	spu->class_0_pending = 0;
-	spu->dar = ea;
-	spu->dsisr = dsisr;
+	spu->class_1_dar = ea;
+	spu->class_1_dsisr = dsisr;
+
+	spu->stop_callback(spu, 1);
 
-	spu->stop_callback(spu);
+	spu->class_1_dar = 0;
+	spu->class_1_dsisr = 0;
 
 	return 0;
 }
@@ -318,11 +320,15 @@ spu_irq_class_0(int irq, void *data)
 	stat = spu_int_stat_get(spu, 0) & mask;
 
 	spu->class_0_pending |= stat;
-	spu->dsisr = spu_mfc_dsisr_get(spu);
-	spu->dar = spu_mfc_dar_get(spu);
+	spu->class_0_dsisr = spu_mfc_dsisr_get(spu);
+	spu->class_0_dar = spu_mfc_dar_get(spu);
 	spin_unlock(&spu->register_lock);
 
-	spu->stop_callback(spu);
+	spu->stop_callback(spu, 0);
+
+	spu->class_0_pending = 0;
+	spu->class_0_dsisr = 0;
+	spu->class_0_dar = 0;
 
 	spu_int_stat_clear(spu, 0, stat);
 
@@ -363,6 +369,9 @@ spu_irq_class_1(int irq, void *data)
 	if (stat & CLASS1_LS_COMPARE_SUSPEND_ON_PUT_INTR)
 		;
 
+	spu->class_1_dsisr = 0;
+	spu->class_1_dar = 0;
+
 	return stat ? IRQ_HANDLED : IRQ_NONE;
 }
 
@@ -396,10 +405,10 @@ spu_irq_class_2(int irq, void *data)
 		spu->ibox_callback(spu);
 
 	if (stat & CLASS2_SPU_STOP_INTR)
-		spu->stop_callback(spu);
+		spu->stop_callback(spu, 2);
 
 	if (stat & CLASS2_SPU_HALT_INTR)
-		spu->stop_callback(spu);
+		spu->stop_callback(spu, 2);
 
 	if (stat & CLASS2_SPU_DMA_TAG_GROUP_COMPLETE_INTR)
 		spu->mfc_callback(spu);
diff --git a/arch/powerpc/platforms/cell/spufs/fault.c b/arch/powerpc/platforms/cell/spufs/fault.c
index e46d300e21a..f093a581ac7 100644
--- a/arch/powerpc/platforms/cell/spufs/fault.c
+++ b/arch/powerpc/platforms/cell/spufs/fault.c
@@ -83,13 +83,18 @@ int spufs_handle_class0(struct spu_context *ctx)
 		return 0;
 
 	if (stat & CLASS0_DMA_ALIGNMENT_INTR)
-		spufs_handle_event(ctx, ctx->csa.dar, SPE_EVENT_DMA_ALIGNMENT);
+		spufs_handle_event(ctx, ctx->csa.class_0_dar,
+			SPE_EVENT_DMA_ALIGNMENT);
 
 	if (stat & CLASS0_INVALID_DMA_COMMAND_INTR)
-		spufs_handle_event(ctx, ctx->csa.dar, SPE_EVENT_INVALID_DMA);
+		spufs_handle_event(ctx, ctx->csa.class_0_dar,
+			SPE_EVENT_INVALID_DMA);
 
 	if (stat & CLASS0_SPU_ERROR_INTR)
-		spufs_handle_event(ctx, ctx->csa.dar, SPE_EVENT_SPE_ERROR);
+		spufs_handle_event(ctx, ctx->csa.class_0_dar,
+			SPE_EVENT_SPE_ERROR);
+
+	ctx->csa.class_0_pending = 0;
 
 	return -EIO;
 }
@@ -119,8 +124,8 @@ int spufs_handle_class1(struct spu_context *ctx)
 	 * in time, we can still expect to get the same fault
 	 * the immediately after the context restore.
 	 */
-	ea = ctx->csa.dar;
-	dsisr = ctx->csa.dsisr;
+	ea = ctx->csa.class_1_dar;
+	dsisr = ctx->csa.class_1_dsisr;
 
 	if (!(dsisr & (MFC_DSISR_PTE_NOT_FOUND | MFC_DSISR_ACCESS_DENIED)))
 		return 0;
@@ -158,7 +163,7 @@ int spufs_handle_class1(struct spu_context *ctx)
 	 * time slicing will not preempt the context while the page fault
 	 * handler is running. Context switch code removes mappings.
 	 */
-	ctx->csa.dar = ctx->csa.dsisr = 0;
+	ctx->csa.class_1_dar = ctx->csa.class_1_dsisr = 0;
 
 	/*
 	 * If we handled the fault successfully and are in runnable
diff --git a/arch/powerpc/platforms/cell/spufs/run.c b/arch/powerpc/platforms/cell/spufs/run.c
index e764a43b544..b7493b86581 100644
--- a/arch/powerpc/platforms/cell/spufs/run.c
+++ b/arch/powerpc/platforms/cell/spufs/run.c
@@ -11,7 +11,7 @@
 #include "spufs.h"
 
 /* interrupt-level stop callback function. */
-void spufs_stop_callback(struct spu *spu)
+void spufs_stop_callback(struct spu *spu, int irq)
 {
 	struct spu_context *ctx = spu->ctx;
 
@@ -24,9 +24,19 @@ void spufs_stop_callback(struct spu *spu)
 	 */
 	if (ctx) {
 		/* Copy exception arguments into module specific structure */
-		ctx->csa.class_0_pending = spu->class_0_pending;
-		ctx->csa.dsisr = spu->dsisr;
-		ctx->csa.dar = spu->dar;
+		switch(irq) {
+		case 0 :
+			ctx->csa.class_0_pending = spu->class_0_pending;
+			ctx->csa.class_0_dsisr = spu->class_0_dsisr;
+			ctx->csa.class_0_dar = spu->class_0_dar;
+			break;
+		case 1 :
+			ctx->csa.class_1_dsisr = spu->class_1_dsisr;
+			ctx->csa.class_1_dar = spu->class_1_dar;
+			break;
+		case 2 :
+			break;
+		}
 
 		/* ensure that the exception status has hit memory before a
 		 * thread waiting on the context's stop queue is woken */
@@ -34,11 +44,6 @@ void spufs_stop_callback(struct spu *spu)
 
 		wake_up_all(&ctx->stop_wq);
 	}
-
-	/* Clear callback arguments from spu structure */
-	spu->class_0_pending = 0;
-	spu->dsisr = 0;
-	spu->dar = 0;
 }
 
 int spu_stopped(struct spu_context *ctx, u32 *stat)
@@ -56,7 +61,11 @@ int spu_stopped(struct spu_context *ctx, u32 *stat)
 	if (!(*stat & SPU_STATUS_RUNNING) && (*stat & stopped))
 		return 1;
 
-	dsisr = ctx->csa.dsisr;
+	dsisr = ctx->csa.class_0_dsisr;
+	if (dsisr & (MFC_DSISR_PTE_NOT_FOUND | MFC_DSISR_ACCESS_DENIED))
+		return 1;
+
+	dsisr = ctx->csa.class_1_dsisr;
 	if (dsisr & (MFC_DSISR_PTE_NOT_FOUND | MFC_DSISR_ACCESS_DENIED))
 		return 1;
 
diff --git a/arch/powerpc/platforms/cell/spufs/spufs.h b/arch/powerpc/platforms/cell/spufs/spufs.h
index dc3a215a6a2..454c277c145 100644
--- a/arch/powerpc/platforms/cell/spufs/spufs.h
+++ b/arch/powerpc/platforms/cell/spufs/spufs.h
@@ -332,7 +332,7 @@ size_t spu_ibox_read(struct spu_context *ctx, u32 *data);
 /* irq callback funcs. */
 void spufs_ibox_callback(struct spu *spu);
 void spufs_wbox_callback(struct spu *spu);
-void spufs_stop_callback(struct spu *spu);
+void spufs_stop_callback(struct spu *spu, int irq);
 void spufs_mfc_callback(struct spu *spu);
 void spufs_dma_callback(struct spu *spu, int type);
 
-- 
cgit v1.2.3


From de1028927ae3487e2e450dacf50fbf32042aee18 Mon Sep 17 00:00:00 2001
From: Luke Browning <lukebr@linux.vnet.ibm.com>
Date: Mon, 28 Apr 2008 17:35:56 +1000
Subject: [POWERPC] spufs: handle faults while the context switch pending flag
 is set

Currently, page fault handlers don't issue a mfc restart if the context
switch pending flag is set, which can leave us with a hanging DMA after
a context restore.

This patch introduces fault pending flag that is set by the fault
handler and read by the context switch code, so that the latter can add
the restart bit at the right spot, after it has successfuly saved the
state of the mfc control register.

Signed-off-by: Luke Browning <lukebr@linux.vnet.ibm.com>
Signed-off-by: Jeremy Kerr <jk@ozlabs.org>
---
 arch/powerpc/platforms/cell/spu_base.c     |  4 ++++
 arch/powerpc/platforms/cell/spufs/switch.c | 23 ++++++++++++++---------
 2 files changed, 18 insertions(+), 9 deletions(-)

(limited to 'arch/powerpc/platforms')

diff --git a/arch/powerpc/platforms/cell/spu_base.c b/arch/powerpc/platforms/cell/spu_base.c
index b9ae675640d..70c660121ec 100644
--- a/arch/powerpc/platforms/cell/spu_base.c
+++ b/arch/powerpc/platforms/cell/spu_base.c
@@ -141,6 +141,10 @@ static void spu_restart_dma(struct spu *spu)
 
 	if (!test_bit(SPU_CONTEXT_SWITCH_PENDING, &spu->flags))
 		out_be64(&priv2->mfc_control_RW, MFC_CNTL_RESTART_DMA_COMMAND);
+	else {
+		set_bit(SPU_CONTEXT_FAULT_PENDING, &spu->flags);
+		mb();
+	}
 }
 
 static inline void spu_load_slb(struct spu *spu, int slbe, struct spu_slb *slb)
diff --git a/arch/powerpc/platforms/cell/spufs/switch.c b/arch/powerpc/platforms/cell/spufs/switch.c
index 47c658051fc..3df9a36eb2f 100644
--- a/arch/powerpc/platforms/cell/spufs/switch.c
+++ b/arch/powerpc/platforms/cell/spufs/switch.c
@@ -139,6 +139,7 @@ static inline void disable_interrupts(struct spu_state *csa, struct spu *spu)
 	 * via a simple load.
 	 */
 	set_bit(SPU_CONTEXT_SWITCH_PENDING, &spu->flags);
+	clear_bit(SPU_CONTEXT_FAULT_PENDING, &spu->flags);
 	synchronize_irq(spu->irqs[0]);
 	synchronize_irq(spu->irqs[1]);
 	synchronize_irq(spu->irqs[2]);
@@ -739,10 +740,14 @@ static inline void set_switch_active(struct spu_state *csa, struct spu *spu)
 	/* Save, Step 48:
 	 * Restore, Step 23.
 	 *     Change the software context switch pending flag
-	 *     to context switch active.
+	 *     to context switch active.  This implementation does
+	 *     not uses a switch active flag.
 	 *
-	 *     This implementation does not uses a switch active flag.
+	 * Now that we have saved the mfc in the csa, we can add in the
+	 * restart command if an exception occurred.
 	 */
+	if (test_bit(SPU_CONTEXT_FAULT_PENDING, &spu->flags))
+		csa->priv2.mfc_control_RW |= MFC_CNTL_RESTART_DMA_COMMAND;
 	clear_bit(SPU_CONTEXT_SWITCH_PENDING, &spu->flags);
 	mb();
 }
@@ -1742,15 +1747,15 @@ static inline void restore_mfc_cntl(struct spu_state *csa, struct spu *spu)
 	 */
 	out_be64(&priv2->mfc_control_RW, csa->priv2.mfc_control_RW);
 	eieio();
+
 	/*
-	 * FIXME: this is to restart a DMA that we were processing
-	 *        before the save. better remember the fault information
-	 *        in the csa instead.
+	 * The queue is put back into the same state that was evident prior to
+	 * the context switch. The suspend flag is added to the saved state in
+	 * the csa, if the operational state was suspending or suspended. In
+	 * this case, the code that suspended the mfc is responsible for
+	 * continuing it. Note that SPE faults do not change the operational
+	 * state of the spu.
 	 */
-	if ((csa->priv2.mfc_control_RW & MFC_CNTL_SUSPEND_DMA_QUEUE_MASK)) {
-		out_be64(&priv2->mfc_control_RW, MFC_CNTL_RESTART_DMA_COMMAND);
-		eieio();
-	}
 }
 
 static inline void enable_user_access(struct spu_state *csa, struct spu *spu)
-- 
cgit v1.2.3


From 826be063eee9b4c2703fd86cfc9723bc391ff1cb Mon Sep 17 00:00:00 2001
From: Christoph Hellwig <hch@lst.de>
Date: Tue, 6 May 2008 09:24:24 +1000
Subject: [POWERPC] spufs: spu_create should send inotify IM_CREATE event

Creating a spufs context or gand using spu_create should send an inotify
event so that things like performance monitors have an easy way to find
out about newly created contexts.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jeremy Kerr <jk@ozlabs.org>
---
 arch/powerpc/platforms/cell/spufs/inode.c | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)

(limited to 'arch/powerpc/platforms')

diff --git a/arch/powerpc/platforms/cell/spufs/inode.c b/arch/powerpc/platforms/cell/spufs/inode.c
index 0c32a05ab06..5faedf5a09a 100644
--- a/arch/powerpc/platforms/cell/spufs/inode.c
+++ b/arch/powerpc/platforms/cell/spufs/inode.c
@@ -23,6 +23,7 @@
 
 #include <linux/file.h>
 #include <linux/fs.h>
+#include <linux/fsnotify.h>
 #include <linux/backing-dev.h>
 #include <linux/init.h>
 #include <linux/ioctl.h>
@@ -618,12 +619,15 @@ long spufs_create(struct nameidata *nd, unsigned int flags, mode_t mode,
 	mode &= ~current->fs->umask;
 
 	if (flags & SPU_CREATE_GANG)
-		return spufs_create_gang(nd->path.dentry->d_inode,
+		ret = spufs_create_gang(nd->path.dentry->d_inode,
 					 dentry, nd->path.mnt, mode);
 	else
-		return spufs_create_context(nd->path.dentry->d_inode,
+		ret = spufs_create_context(nd->path.dentry->d_inode,
 					    dentry, nd->path.mnt, flags, mode,
 					    filp);
+	if (ret >= 0)
+		fsnotify_mkdir(nd->path.dentry->d_inode, dentry);
+	return ret;
 
 out_dput:
 	dput(dentry);
-- 
cgit v1.2.3


From 7a28a1549f9514f3b0dd3dde5c7337ba5d44fba3 Mon Sep 17 00:00:00 2001
From: Christoph Hellwig <hch@lst.de>
Date: Thu, 8 May 2008 15:26:32 +1000
Subject: [POWERPC] spufs: don't requeue victim contex in find_victim if it's
 not in spu_run

We should not requeue the victim context in find_victim if the owner is
not in spu_run. It's first not needed because leaving the context on
the spu is an optimization and second is harmful because it means the
owner could re-enter spu_run when the context is on the runqueue and
trip the BUG_ON in __spu_update_sched_info.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jeremy Kerr <jk@ozlabs.org>
---
 arch/powerpc/platforms/cell/spufs/sched.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

(limited to 'arch/powerpc/platforms')

diff --git a/arch/powerpc/platforms/cell/spufs/sched.c b/arch/powerpc/platforms/cell/spufs/sched.c
index 1df7d6d152c..8a05ac863a2 100644
--- a/arch/powerpc/platforms/cell/spufs/sched.c
+++ b/arch/powerpc/platforms/cell/spufs/sched.c
@@ -659,7 +659,8 @@ static struct spu *find_victim(struct spu_context *ctx)
 
 			victim->stats.invol_ctx_switch++;
 			spu->stats.invol_ctx_switch++;
-			spu_add_to_rq(victim);
+			if (test_bit(SPU_SCHED_SPU_RUN, &ctx->sched_flags))
+				spu_add_to_rq(victim);
 
 			mutex_unlock(&victim->state_mutex);
 
-- 
cgit v1.2.3


From 02539d71fa98d5737bb668b02286c76241e4bac9 Mon Sep 17 00:00:00 2001
From: Christoph Hellwig <hch@lst.de>
Date: Thu, 8 May 2008 15:29:12 +1000
Subject: [POWERPC] spufs: lockdep annotations for spufs_dir_close

We need to acquire the parent i_mutex with I_MUTEX_PARENT to keep
lockdep happy.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jeremy Kerr <jk@ozlabs.org>
---
 arch/powerpc/platforms/cell/spufs/inode.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

(limited to 'arch/powerpc/platforms')

diff --git a/arch/powerpc/platforms/cell/spufs/inode.c b/arch/powerpc/platforms/cell/spufs/inode.c
index 5faedf5a09a..f407b247185 100644
--- a/arch/powerpc/platforms/cell/spufs/inode.c
+++ b/arch/powerpc/platforms/cell/spufs/inode.c
@@ -224,7 +224,7 @@ static int spufs_dir_close(struct inode *inode, struct file *file)
 	parent = dir->d_parent->d_inode;
 	ctx = SPUFS_I(dir->d_inode)->i_ctx;
 
-	mutex_lock(&parent->i_mutex);
+	mutex_lock_nested(&parent->i_mutex, I_MUTEX_PARENT);
 	ret = spufs_rmdir(parent, dir);
 	mutex_unlock(&parent->i_mutex);
 	WARN_ON(ret);
-- 
cgit v1.2.3