aboutsummaryrefslogtreecommitdiff
path: root/arch/powerpc/platforms/cell
diff options
context:
space:
mode:
Diffstat (limited to 'arch/powerpc/platforms/cell')
-rw-r--r--arch/powerpc/platforms/cell/celleb_scc_pciex.c5
-rw-r--r--arch/powerpc/platforms/cell/interrupt.c53
-rw-r--r--arch/powerpc/platforms/cell/io-workarounds.c6
-rw-r--r--arch/powerpc/platforms/cell/io-workarounds.h6
-rw-r--r--arch/powerpc/platforms/cell/spu_base.c69
-rw-r--r--arch/powerpc/platforms/cell/spu_priv1_mmio.c16
-rw-r--r--arch/powerpc/platforms/cell/spufs/coredump.c1
-rw-r--r--arch/powerpc/platforms/cell/spufs/fault.c17
-rw-r--r--arch/powerpc/platforms/cell/spufs/file.c1
-rw-r--r--arch/powerpc/platforms/cell/spufs/inode.c10
-rw-r--r--arch/powerpc/platforms/cell/spufs/run.c51
-rw-r--r--arch/powerpc/platforms/cell/spufs/sched.c26
-rw-r--r--arch/powerpc/platforms/cell/spufs/spufs.h3
-rw-r--r--arch/powerpc/platforms/cell/spufs/switch.c71
14 files changed, 242 insertions, 93 deletions
diff --git a/arch/powerpc/platforms/cell/celleb_scc_pciex.c b/arch/powerpc/platforms/cell/celleb_scc_pciex.c
index 31da84c458d..0e04f8fb152 100644
--- a/arch/powerpc/platforms/cell/celleb_scc_pciex.c
+++ b/arch/powerpc/platforms/cell/celleb_scc_pciex.c
@@ -217,7 +217,7 @@ static u##size scc_pciex_in##name(unsigned long port) \
static void scc_pciex_ins##name(unsigned long p, void *b, unsigned long c) \
{ \
struct iowa_bus *bus = iowa_pio_find_bus(p); \
- u##size *dst = b; \
+ __le##size *dst = b; \
for (; c != 0; c--, dst++) \
*dst = cpu_to_le##size(__scc_pciex_in##name(bus->phb, p)); \
scc_pciex_io_flush(bus); \
@@ -231,10 +231,11 @@ static void scc_pciex_outs##name(unsigned long p, const void *b, \
unsigned long c) \
{ \
struct iowa_bus *bus = iowa_pio_find_bus(p); \
- const u##size *src = b; \
+ const __le##size *src = b; \
for (; c != 0; c--, src++) \
__scc_pciex_out##name(bus->phb, le##size##_to_cpu(*src), p); \
}
+#define __le8 u8
#define cpu_to_le8(x) (x)
#define le8_to_cpu(x) (x)
PCIEX_PIO_FUNC(8, b)
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;
}
diff --git a/arch/powerpc/platforms/cell/io-workarounds.c b/arch/powerpc/platforms/cell/io-workarounds.c
index 3b84e8be314..b5f84e8f089 100644
--- a/arch/powerpc/platforms/cell/io-workarounds.c
+++ b/arch/powerpc/platforms/cell/io-workarounds.c
@@ -118,7 +118,7 @@ static void iowa_##name at \
#undef DEF_PCI_AC_RET
#undef DEF_PCI_AC_NORET
-static struct ppc_pci_io __initdata iowa_pci_io = {
+static const struct ppc_pci_io __devinitconst iowa_pci_io = {
#define DEF_PCI_AC_RET(name, ret, at, al, space, aa) .name = iowa_##name,
#define DEF_PCI_AC_NORET(name, at, al, space, aa) .name = iowa_##name,
@@ -146,7 +146,7 @@ static void __iomem *iowa_ioremap(unsigned long addr, unsigned long size,
}
/* Regist new bus to support workaround */
-void __init iowa_register_bus(struct pci_controller *phb,
+void __devinit iowa_register_bus(struct pci_controller *phb,
struct ppc_pci_io *ops,
int (*initfunc)(struct iowa_bus *, void *), void *data)
{
@@ -173,7 +173,7 @@ void __init iowa_register_bus(struct pci_controller *phb,
}
/* enable IO workaround */
-void __init io_workaround_init(void)
+void __devinit io_workaround_init(void)
{
static int io_workaround_inited;
diff --git a/arch/powerpc/platforms/cell/io-workarounds.h b/arch/powerpc/platforms/cell/io-workarounds.h
index 79d8ed3d510..6efc7782ebf 100644
--- a/arch/powerpc/platforms/cell/io-workarounds.h
+++ b/arch/powerpc/platforms/cell/io-workarounds.h
@@ -31,9 +31,9 @@ struct iowa_bus {
void *private;
};
-void __init io_workaround_init(void);
-void __init iowa_register_bus(struct pci_controller *, struct ppc_pci_io *,
- int (*)(struct iowa_bus *, void *), void *);
+void __devinit io_workaround_init(void);
+void __devinit iowa_register_bus(struct pci_controller *, struct ppc_pci_io *,
+ int (*)(struct iowa_bus *, void *), void *);
struct iowa_bus *iowa_mem_find_bus(const PCI_IO_ADDR);
struct iowa_bus *iowa_pio_find_bus(unsigned long);
diff --git a/arch/powerpc/platforms/cell/spu_base.c b/arch/powerpc/platforms/cell/spu_base.c
index 6bab44b7716..78f905bc6a4 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)
@@ -215,22 +219,34 @@ static int __spu_trap_data_seg(struct spu *spu, unsigned long ea)
extern int hash_page(unsigned long ea, unsigned long access, unsigned long trap); //XXX
static int __spu_trap_data_map(struct spu *spu, unsigned long ea, u64 dsisr)
{
+ int ret;
+
pr_debug("%s, %lx, %lx\n", __func__, dsisr, ea);
- /* Handle kernel space hash faults immediately.
- User hash faults need to be deferred to process context. */
- if ((dsisr & MFC_DSISR_PTE_NOT_FOUND)
- && REGION_ID(ea) != USER_REGION_ID
- && hash_page(ea, _PAGE_PRESENT, 0x300) == 0) {
- spu_restart_dma(spu);
- return 0;
+ /*
+ * Handle kernel space hash faults immediately. User hash
+ * faults need to be deferred to process context.
+ */
+ if ((dsisr & MFC_DSISR_PTE_NOT_FOUND) &&
+ (REGION_ID(ea) != USER_REGION_ID)) {
+
+ spin_unlock(&spu->register_lock);
+ ret = hash_page(ea, _PAGE_PRESENT, 0x300);
+ spin_lock(&spu->register_lock);
+
+ if (!ret) {
+ spu_restart_dma(spu);
+ 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;
}
@@ -318,13 +334,13 @@ 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);
- spin_unlock(&spu->register_lock);
-
- spu->stop_callback(spu);
+ spu->class_0_dar = spu_mfc_dar_get(spu);
+ spu->stop_callback(spu, 0);
+ spu->class_0_pending = 0;
+ spu->class_0_dar = 0;
spu_int_stat_clear(spu, 0, stat);
+ spin_unlock(&spu->register_lock);
return IRQ_HANDLED;
}
@@ -347,13 +363,12 @@ spu_irq_class_1(int irq, void *data)
spu_mfc_dsisr_set(spu, 0ul);
spu_int_stat_clear(spu, 1, stat);
- if (stat & CLASS1_SEGMENT_FAULT_INTR)
- __spu_trap_data_seg(spu, dar);
-
- spin_unlock(&spu->register_lock);
pr_debug("%s: %lx %lx %lx %lx\n", __func__, mask, stat,
dar, dsisr);
+ if (stat & CLASS1_SEGMENT_FAULT_INTR)
+ __spu_trap_data_seg(spu, dar);
+
if (stat & CLASS1_STORAGE_FAULT_INTR)
__spu_trap_data_map(spu, dar, dsisr);
@@ -363,6 +378,11 @@ 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;
+
+ spin_unlock(&spu->register_lock);
+
return stat ? IRQ_HANDLED : IRQ_NONE;
}
@@ -381,14 +401,12 @@ spu_irq_class_2(int irq, void *data)
mask = spu_int_mask_get(spu, 2);
/* ignore interrupts we're not waiting for */
stat &= mask;
-
/* mailbox interrupts are level triggered. mask them now before
* acknowledging */
if (stat & mailbox_intrs)
spu_int_mask_and(spu, 2, ~(stat & mailbox_intrs));
/* acknowledge all interrupts before the callbacks */
spu_int_stat_clear(spu, 2, stat);
- spin_unlock(&spu->register_lock);
pr_debug("class 2 interrupt %d, %lx, %lx\n", irq, stat, mask);
@@ -396,10 +414,10 @@ 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);
@@ -408,6 +426,9 @@ spu_irq_class_2(int irq, void *data)
spu->wbox_callback(spu);
spu->stats.class2_intr++;
+
+ spin_unlock(&spu->register_lock);
+
return stat ? IRQ_HANDLED : IRQ_NONE;
}
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/coredump.c b/arch/powerpc/platforms/cell/spufs/coredump.c
index b962c3ab470..af116aadba1 100644
--- a/arch/powerpc/platforms/cell/spufs/coredump.c
+++ b/arch/powerpc/platforms/cell/spufs/coredump.c
@@ -22,6 +22,7 @@
#include <linux/elf.h>
#include <linux/file.h>
+#include <linux/fdtable.h>
#include <linux/fs.h>
#include <linux/list.h>
#include <linux/module.h>
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/file.c b/arch/powerpc/platforms/cell/spufs/file.c
index 80911a37340..c81341ff75b 100644
--- a/arch/powerpc/platforms/cell/spufs/file.c
+++ b/arch/powerpc/platforms/cell/spufs/file.c
@@ -32,6 +32,7 @@
#include <linux/marker.h>
#include <asm/io.h>
+#include <asm/time.h>
#include <asm/spu.h>
#include <asm/spu_info.h>
#include <asm/uaccess.h>
diff --git a/arch/powerpc/platforms/cell/spufs/inode.c b/arch/powerpc/platforms/cell/spufs/inode.c
index 0c32a05ab06..f407b247185 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>
@@ -223,7 +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);
@@ -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);
diff --git a/arch/powerpc/platforms/cell/spufs/run.c b/arch/powerpc/platforms/cell/spufs/run.c
index a9c35b7b719..f7edba6cb79 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 +24,18 @@ 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_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 +43,6 @@ 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)
@@ -46,17 +50,25 @@ int spu_stopped(struct spu_context *ctx, u32 *stat)
u64 dsisr;
u32 stopped;
- *stat = ctx->ops->status_read(ctx);
+ stopped = SPU_STATUS_INVALID_INSTR | SPU_STATUS_SINGLE_STEP |
+ SPU_STATUS_STOPPED_BY_HALT | SPU_STATUS_STOPPED_BY_STOP;
- if (test_bit(SPU_SCHED_NOTIFY_ACTIVE, &ctx->sched_flags))
+top:
+ *stat = ctx->ops->status_read(ctx);
+ if (*stat & stopped) {
+ /*
+ * If the spu hasn't finished stopping, we need to
+ * re-read the register to get the stopped value.
+ */
+ if (*stat & SPU_STATUS_RUNNING)
+ goto top;
return 1;
+ }
- stopped = SPU_STATUS_INVALID_INSTR | SPU_STATUS_SINGLE_STEP |
- SPU_STATUS_STOPPED_BY_HALT | SPU_STATUS_STOPPED_BY_STOP;
- if (!(*stat & SPU_STATUS_RUNNING) && (*stat & stopped))
+ if (test_bit(SPU_SCHED_NOTIFY_ACTIVE, &ctx->sched_flags))
return 1;
- dsisr = ctx->csa.dsisr;
+ dsisr = ctx->csa.class_1_dsisr;
if (dsisr & (MFC_DSISR_PTE_NOT_FOUND | MFC_DSISR_ACCESS_DENIED))
return 1;
@@ -294,7 +306,7 @@ 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 +328,9 @@ 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
@@ -343,13 +353,14 @@ 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);
diff --git a/arch/powerpc/platforms/cell/spufs/sched.c b/arch/powerpc/platforms/cell/spufs/sched.c
index 7298e7db2c8..e929e70a84e 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)
@@ -227,23 +230,26 @@ static void spu_bind_context(struct spu *spu, struct spu_context *ctx)
ctx->stats.slb_flt_base = spu->stats.slb_flt;
ctx->stats.class2_intr_base = spu->stats.class2_intr;
+ spu_associate_mm(spu, ctx->owner);
+
+ spin_lock_irq(&spu->register_lock);
spu->ctx = ctx;
spu->flags = 0;
ctx->spu = spu;
ctx->ops = &spu_hw_ops;
spu->pid = current->pid;
spu->tgid = current->tgid;
- spu_associate_mm(spu, ctx->owner);
spu->ibox_callback = spufs_ibox_callback;
spu->wbox_callback = spufs_wbox_callback;
spu->stop_callback = spufs_stop_callback;
spu->mfc_callback = spufs_mfc_callback;
- mb();
+ spin_unlock_irq(&spu->register_lock);
+
spu_unmap_mappings(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;
@@ -401,6 +407,8 @@ static int has_affinity(struct spu_context *ctx)
*/
static void spu_unbind_context(struct spu *spu, struct spu_context *ctx)
{
+ u32 status;
+
spu_context_trace(spu_unbind_context__enter, ctx, spu);
spuctx_switch_state(ctx, SPU_UTIL_SYSTEM);
@@ -421,18 +429,22 @@ static void spu_unbind_context(struct spu *spu, struct spu_context *ctx)
spu_unmap_mappings(ctx);
spu_save(&ctx->csa, spu);
spu_switch_log_notify(spu, ctx, SWITCH_LOG_STOP, 0);
+
+ spin_lock_irq(&spu->register_lock);
spu->timestamp = jiffies;
ctx->state = SPU_STATE_SAVED;
spu->ibox_callback = NULL;
spu->wbox_callback = NULL;
spu->stop_callback = NULL;
spu->mfc_callback = NULL;
- spu_associate_mm(spu, NULL);
spu->pid = 0;
spu->tgid = 0;
ctx->ops = &spu_backing_ops;
spu->flags = 0;
spu->ctx = NULL;
+ spin_unlock_irq(&spu->register_lock);
+
+ spu_associate_mm(spu, NULL);
ctx->stats.slb_flt +=
(spu->stats.slb_flt - ctx->stats.slb_flt_base);
@@ -442,6 +454,9 @@ static void spu_unbind_context(struct spu *spu, struct spu_context *ctx)
/* This maps the underlying spu state to idle */
spuctx_switch_state(ctx, SPU_UTIL_IDLE_LOADED);
ctx->spu = NULL;
+
+ if (spu_stopped(ctx, &status))
+ wake_up_all(&ctx->stop_wq);
}
/**
@@ -657,7 +672,8 @@ 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, &victim->sched_flags))
+ spu_add_to_rq(victim);
mutex_unlock(&victim->state_mutex);
diff --git a/arch/powerpc/platforms/cell/spufs/spufs.h b/arch/powerpc/platforms/cell/spufs/spufs.h
index 7312745b754..454c277c145 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 {
@@ -331,7 +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);
diff --git a/arch/powerpc/platforms/cell/spufs/switch.c b/arch/powerpc/platforms/cell/spufs/switch.c
index d2a1249d36d..3df9a36eb2f 100644
--- a/arch/powerpc/platforms/cell/spufs/switch.c
+++ b/arch/powerpc/platforms/cell/spufs/switch.c
@@ -132,6 +132,14 @@ 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);
+ clear_bit(SPU_CONTEXT_FAULT_PENDING, &spu->flags);
synchronize_irq(spu->irqs[0]);
synchronize_irq(spu->irqs[1]);
synchronize_irq(spu->irqs[2]);
@@ -166,9 +174,8 @@ 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)
@@ -186,20 +193,21 @@ 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;
}
}
@@ -249,16 +257,21 @@ 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)
@@ -462,7 +475,9 @@ 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();
}
@@ -725,10 +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();
}
@@ -1690,6 +1709,13 @@ 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)
{
@@ -1721,15 +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)
@@ -1788,7 +1814,7 @@ 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. */
@@ -2000,6 +2026,7 @@ 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. */
generated by cgit v1.2.3 (git 2.25.1) at 2024-07-15 00:24:21 +0000