* Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
*/
-#define DEBUG 1
+#undef DEBUG
#include <linux/interrupt.h>
#include <linux/list.h>
#include <linux/module.h>
-#include <linux/poll.h>
#include <linux/ptrace.h>
#include <linux/slab.h>
#include <linux/wait.h>
-
-#include <asm/io.h>
-#include <asm/prom.h>
-#include <asm/semaphore.h>
+#include <linux/mm.h>
+#include <linux/io.h>
+#include <linux/mutex.h>
+#include <linux/linux_logo.h>
#include <asm/spu.h>
-#include <asm/mmu_context.h>
+#include <asm/spu_priv1.h>
+#include <asm/xmon.h>
+
+const struct spu_management_ops *spu_management_ops;
+EXPORT_SYMBOL_GPL(spu_management_ops);
+
+const struct spu_priv1_ops *spu_priv1_ops;
+
+static struct list_head spu_list[MAX_NUMNODES];
+static LIST_HEAD(spu_full_list);
+static DEFINE_MUTEX(spu_mutex);
+static DEFINE_SPINLOCK(spu_list_lock);
+
+EXPORT_SYMBOL_GPL(spu_priv1_ops);
+
+void spu_invalidate_slbs(struct spu *spu)
+{
+ struct spu_priv2 __iomem *priv2 = spu->priv2;
-#include "interrupt.h"
+ if (spu_mfc_sr1_get(spu) & MFC_STATE1_RELOCATE_MASK)
+ out_be64(&priv2->slb_invalidate_all_W, 0UL);
+}
+EXPORT_SYMBOL_GPL(spu_invalidate_slbs);
+
+/* This is called by the MM core when a segment size is changed, to
+ * request a flush of all the SPEs using a given mm
+ */
+void spu_flush_all_slbs(struct mm_struct *mm)
+{
+ struct spu *spu;
+ unsigned long flags;
+
+ spin_lock_irqsave(&spu_list_lock, flags);
+ list_for_each_entry(spu, &spu_full_list, full_list) {
+ if (spu->mm == mm)
+ spu_invalidate_slbs(spu);
+ }
+ spin_unlock_irqrestore(&spu_list_lock, flags);
+}
+
+/* The hack below stinks... try to do something better one of
+ * these days... Does it even work properly with NR_CPUS == 1 ?
+ */
+static inline void mm_needs_global_tlbie(struct mm_struct *mm)
+{
+ int nr = (NR_CPUS > 1) ? NR_CPUS : NR_CPUS + 1;
+
+ /* Global TLBIE broadcast required with SPEs. */
+ __cpus_setall(&mm->cpu_vm_mask, nr);
+}
+
+void spu_associate_mm(struct spu *spu, struct mm_struct *mm)
+{
+ unsigned long flags;
+
+ spin_lock_irqsave(&spu_list_lock, flags);
+ spu->mm = mm;
+ spin_unlock_irqrestore(&spu_list_lock, flags);
+ if (mm)
+ mm_needs_global_tlbie(mm);
+}
+EXPORT_SYMBOL_GPL(spu_associate_mm);
static int __spu_trap_invalid_dma(struct spu *spu)
{
pr_debug("%s\n", __FUNCTION__);
- force_sig(SIGBUS, /* info, */ current);
+ spu->dma_callback(spu, SPE_EVENT_INVALID_DMA);
return 0;
}
static int __spu_trap_dma_align(struct spu *spu)
{
pr_debug("%s\n", __FUNCTION__);
- force_sig(SIGBUS, /* info, */ current);
+ spu->dma_callback(spu, SPE_EVENT_DMA_ALIGNMENT);
return 0;
}
static int __spu_trap_error(struct spu *spu)
{
pr_debug("%s\n", __FUNCTION__);
- force_sig(SIGILL, /* info, */ current);
+ spu->dma_callback(spu, SPE_EVENT_SPE_ERROR);
return 0;
}
static void spu_restart_dma(struct spu *spu)
{
struct spu_priv2 __iomem *priv2 = spu->priv2;
- out_be64(&priv2->mfc_control_RW, MFC_CNTL_RESTART_DMA_COMMAND);
+
+ if (!test_bit(SPU_CONTEXT_SWITCH_PENDING, &spu->flags))
+ out_be64(&priv2->mfc_control_RW, MFC_CNTL_RESTART_DMA_COMMAND);
}
static int __spu_trap_data_seg(struct spu *spu, unsigned long ea)
{
- struct spu_priv2 __iomem *priv2;
- struct mm_struct *mm;
+ struct spu_priv2 __iomem *priv2 = spu->priv2;
+ struct mm_struct *mm = spu->mm;
+ u64 esid, vsid, llp;
+ int psize;
pr_debug("%s\n", __FUNCTION__);
- if (REGION_ID(ea) != USER_REGION_ID) {
+ if (test_bit(SPU_CONTEXT_SWITCH_ACTIVE, &spu->flags)) {
+ /* SLBs are pre-loaded for context switch, so
+ * we should never get here!
+ */
+ printk("%s: invalid access during switch!\n", __func__);
+ return 1;
+ }
+ esid = (ea & ESID_MASK) | SLB_ESID_V;
+
+ switch(REGION_ID(ea)) {
+ case USER_REGION_ID:
+#ifdef CONFIG_PPC_MM_SLICES
+ psize = get_slice_psize(mm, ea);
+#else
+ psize = mm->context.user_psize;
+#endif
+ vsid = (get_vsid(mm->context.id, ea) << SLB_VSID_SHIFT) |
+ SLB_VSID_USER;
+ break;
+ case VMALLOC_REGION_ID:
+ if (ea < VMALLOC_END)
+ psize = mmu_vmalloc_psize;
+ else
+ psize = mmu_io_psize;
+ vsid = (get_kernel_vsid(ea) << SLB_VSID_SHIFT) |
+ SLB_VSID_KERNEL;
+ break;
+ case KERNEL_REGION_ID:
+ psize = mmu_linear_psize;
+ vsid = (get_kernel_vsid(ea) << SLB_VSID_SHIFT) |
+ SLB_VSID_KERNEL;
+ break;
+ default:
+ /* Future: support kernel segments so that drivers
+ * can use SPUs.
+ */
pr_debug("invalid region access at %016lx\n", ea);
return 1;
}
+ llp = mmu_psize_defs[psize].sllp;
- priv2 = spu->priv2;
- mm = spu->mm;
+ out_be64(&priv2->slb_index_W, spu->slb_replace);
+ out_be64(&priv2->slb_vsid_RW, vsid | llp);
+ out_be64(&priv2->slb_esid_RW, esid);
+ spu->slb_replace++;
if (spu->slb_replace >= 8)
spu->slb_replace = 0;
- out_be64(&priv2->slb_index_W, spu->slb_replace);
- out_be64(&priv2->slb_vsid_RW,
- (get_vsid(mm->context.id, ea) << SLB_VSID_SHIFT)
- | SLB_VSID_USER);
- out_be64(&priv2->slb_esid_RW, (ea & ESID_MASK) | SLB_ESID_V);
-
spu_restart_dma(spu);
-
- pr_debug("set slb %d context %lx, ea %016lx, vsid %016lx, esid %016lx\n",
- spu->slb_replace, mm->context.id, ea,
- (get_vsid(mm->context.id, ea) << SLB_VSID_SHIFT)| SLB_VSID_USER,
- (ea & ESID_MASK) | SLB_ESID_V);
- return 0;
-}
-
-static int __spu_trap_data_map(struct spu *spu, unsigned long ea)
-{
- unsigned long dsisr;
- struct spu_priv1 __iomem *priv1;
-
- pr_debug("%s\n", __FUNCTION__);
- priv1 = spu->priv1;
- dsisr = in_be64(&priv1->mfc_dsisr_RW);
-
- wake_up(&spu->stop_wq);
-
- return 0;
-}
-
-static int __spu_trap_mailbox(struct spu *spu)
-{
- wake_up_all(&spu->ibox_wq);
- kill_fasync(&spu->ibox_fasync, SIGIO, POLLIN);
-
- /* atomically disable SPU mailbox interrupts */
- spin_lock(&spu->register_lock);
- out_be64(&spu->priv1->int_mask_class2_RW,
- in_be64(&spu->priv1->int_mask_class2_RW) & ~0x1);
- spin_unlock(&spu->register_lock);
- return 0;
-}
-
-static int __spu_trap_stop(struct spu *spu)
-{
- pr_debug("%s\n", __FUNCTION__);
- spu->stop_code = in_be32(&spu->problem->spu_status_R);
- wake_up(&spu->stop_wq);
+ spu->stats.slb_flt++;
return 0;
}
-static int __spu_trap_halt(struct spu *spu)
+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)
{
- pr_debug("%s\n", __FUNCTION__);
- spu->stop_code = in_be32(&spu->problem->spu_status_R);
- wake_up(&spu->stop_wq);
- return 0;
-}
+ pr_debug("%s, %lx, %lx\n", __FUNCTION__, dsisr, ea);
-static int __spu_trap_tag_group(struct spu *spu)
-{
- pr_debug("%s\n", __FUNCTION__);
- /* wake_up(&spu->dma_wq); */
- 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
+ && hash_page(ea, _PAGE_PRESENT, 0x300) == 0) {
+ spu_restart_dma(spu);
+ return 0;
+ }
-static int __spu_trap_spubox(struct spu *spu)
-{
- wake_up_all(&spu->wbox_wq);
- kill_fasync(&spu->wbox_fasync, SIGIO, POLLOUT);
+ if (test_bit(SPU_CONTEXT_SWITCH_ACTIVE, &spu->flags)) {
+ printk("%s: invalid access during switch!\n", __func__);
+ return 1;
+ }
- /* atomically disable SPU mailbox interrupts */
- spin_lock(&spu->register_lock);
- out_be64(&spu->priv1->int_mask_class2_RW,
- in_be64(&spu->priv1->int_mask_class2_RW) & ~0x10);
- spin_unlock(&spu->register_lock);
+ spu->dar = ea;
+ spu->dsisr = dsisr;
+ mb();
+ spu->stop_callback(spu);
return 0;
}
static irqreturn_t
-spu_irq_class_0(int irq, void *data, struct pt_regs *regs)
+spu_irq_class_0(int irq, void *data)
{
struct spu *spu;
spu = data;
spu->class_0_pending = 1;
- wake_up(&spu->stop_wq);
+ spu->stop_callback(spu);
return IRQ_HANDLED;
}
-static int
+int
spu_irq_class_0_bottom(struct spu *spu)
{
- unsigned long stat;
+ unsigned long stat, mask;
+ unsigned long flags;
spu->class_0_pending = 0;
- stat = in_be64(&spu->priv1->int_stat_class0_RW);
+ spin_lock_irqsave(&spu->register_lock, flags);
+ mask = spu_int_mask_get(spu, 0);
+ stat = spu_int_stat_get(spu, 0);
- if (stat & 1) /* invalid MFC DMA */
- __spu_trap_invalid_dma(spu);
+ stat &= mask;
- if (stat & 2) /* invalid DMA alignment */
+ if (stat & 1) /* invalid DMA alignment */
__spu_trap_dma_align(spu);
+ if (stat & 2) /* invalid MFC DMA */
+ __spu_trap_invalid_dma(spu);
+
if (stat & 4) /* error on SPU */
__spu_trap_error(spu);
- out_be64(&spu->priv1->int_stat_class0_RW, stat);
- return 0;
+ spu_int_stat_clear(spu, 0, stat);
+ spin_unlock_irqrestore(&spu->register_lock, flags);
+
+ return (stat & 0x7) ? -EIO : 0;
}
+EXPORT_SYMBOL_GPL(spu_irq_class_0_bottom);
static irqreturn_t
-spu_irq_class_1(int irq, void *data, struct pt_regs *regs)
+spu_irq_class_1(int irq, void *data)
{
struct spu *spu;
- unsigned long stat, dar;
+ unsigned long stat, mask, dar, dsisr;
spu = data;
- stat = in_be64(&spu->priv1->int_stat_class1_RW);
- dar = in_be64(&spu->priv1->mfc_dar_RW);
+
+ /* atomically read & clear class1 status. */
+ spin_lock(&spu->register_lock);
+ mask = spu_int_mask_get(spu, 1);
+ stat = spu_int_stat_get(spu, 1) & mask;
+ dar = spu_mfc_dar_get(spu);
+ dsisr = spu_mfc_dsisr_get(spu);
+ if (stat & 2) /* mapping fault */
+ spu_mfc_dsisr_set(spu, 0ul);
+ spu_int_stat_clear(spu, 1, stat);
+ spin_unlock(&spu->register_lock);
+ pr_debug("%s: %lx %lx %lx %lx\n", __FUNCTION__, mask, stat,
+ dar, dsisr);
if (stat & 1) /* segment fault */
__spu_trap_data_seg(spu, dar);
if (stat & 2) { /* mapping fault */
- __spu_trap_data_map(spu, dar);
+ __spu_trap_data_map(spu, dar, dsisr);
}
if (stat & 4) /* ls compare & suspend on get */
if (stat & 8) /* ls compare & suspend on put */
;
- out_be64(&spu->priv1->int_stat_class1_RW, stat);
return stat ? IRQ_HANDLED : IRQ_NONE;
}
static irqreturn_t
-spu_irq_class_2(int irq, void *data, struct pt_regs *regs)
+spu_irq_class_2(int irq, void *data)
{
struct spu *spu;
unsigned long stat;
+ unsigned long mask;
spu = data;
- stat = in_be64(&spu->priv1->int_stat_class2_RW);
-
- pr_debug("class 2 interrupt %d, %lx, %lx\n", irq, stat,
- in_be64(&spu->priv1->int_mask_class2_RW));
+ spin_lock(&spu->register_lock);
+ stat = spu_int_stat_get(spu, 2);
+ mask = spu_int_mask_get(spu, 2);
+ /* ignore interrupts we're not waiting for */
+ stat &= mask;
+ /*
+ * mailbox interrupts (0x1 and 0x10) are level triggered.
+ * mask them now before acknowledging.
+ */
+ if (stat & 0x11)
+ spu_int_mask_and(spu, 2, ~(stat & 0x11));
+ /* 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);
if (stat & 1) /* PPC core mailbox */
- __spu_trap_mailbox(spu);
+ spu->ibox_callback(spu);
if (stat & 2) /* SPU stop-and-signal */
- __spu_trap_stop(spu);
+ spu->stop_callback(spu);
if (stat & 4) /* SPU halted */
- __spu_trap_halt(spu);
+ spu->stop_callback(spu);
if (stat & 8) /* DMA tag group complete */
- __spu_trap_tag_group(spu);
+ spu->mfc_callback(spu);
if (stat & 0x10) /* SPU mailbox threshold */
- __spu_trap_spubox(spu);
+ spu->wbox_callback(spu);
- out_be64(&spu->priv1->int_stat_class2_RW, stat);
+ spu->stats.class2_intr++;
return stat ? IRQ_HANDLED : IRQ_NONE;
}
-static int
-spu_request_irqs(struct spu *spu)
+static int spu_request_irqs(struct spu *spu)
{
- int ret;
- int irq_base;
-
- irq_base = IIC_NODE_STRIDE * spu->node + IIC_SPE_OFFSET;
-
- snprintf(spu->irq_c0, sizeof (spu->irq_c0), "spe%02d.0", spu->number);
- ret = request_irq(irq_base + spu->isrc,
- spu_irq_class_0, 0, spu->irq_c0, spu);
- if (ret)
- goto out;
- out_be64(&spu->priv1->int_mask_class0_RW, 0x7);
-
- snprintf(spu->irq_c1, sizeof (spu->irq_c1), "spe%02d.1", spu->number);
- ret = request_irq(irq_base + IIC_CLASS_STRIDE + spu->isrc,
- spu_irq_class_1, 0, spu->irq_c1, spu);
- if (ret)
- goto out1;
- out_be64(&spu->priv1->int_mask_class1_RW, 0x3);
+ int ret = 0;
- snprintf(spu->irq_c2, sizeof (spu->irq_c2), "spe%02d.2", spu->number);
- ret = request_irq(irq_base + 2*IIC_CLASS_STRIDE + spu->isrc,
- spu_irq_class_2, 0, spu->irq_c2, spu);
- if (ret)
- goto out2;
- out_be64(&spu->priv1->int_mask_class2_RW, 0xe);
- goto out;
+ if (spu->irqs[0] != NO_IRQ) {
+ snprintf(spu->irq_c0, sizeof (spu->irq_c0), "spe%02d.0",
+ spu->number);
+ ret = request_irq(spu->irqs[0], spu_irq_class_0,
+ IRQF_DISABLED,
+ spu->irq_c0, spu);
+ if (ret)
+ goto bail0;
+ }
+ if (spu->irqs[1] != NO_IRQ) {
+ snprintf(spu->irq_c1, sizeof (spu->irq_c1), "spe%02d.1",
+ spu->number);
+ ret = request_irq(spu->irqs[1], spu_irq_class_1,
+ IRQF_DISABLED,
+ spu->irq_c1, spu);
+ if (ret)
+ goto bail1;
+ }
+ if (spu->irqs[2] != NO_IRQ) {
+ snprintf(spu->irq_c2, sizeof (spu->irq_c2), "spe%02d.2",
+ spu->number);
+ ret = request_irq(spu->irqs[2], spu_irq_class_2,
+ IRQF_DISABLED,
+ spu->irq_c2, spu);
+ if (ret)
+ goto bail2;
+ }
+ return 0;
-out2:
- free_irq(irq_base + IIC_CLASS_STRIDE + spu->isrc, spu);
-out1:
- free_irq(irq_base + spu->isrc, spu);
-out:
+bail2:
+ if (spu->irqs[1] != NO_IRQ)
+ free_irq(spu->irqs[1], spu);
+bail1:
+ if (spu->irqs[0] != NO_IRQ)
+ free_irq(spu->irqs[0], spu);
+bail0:
return ret;
}
-static void
-spu_free_irqs(struct spu *spu)
+static void spu_free_irqs(struct spu *spu)
{
- int irq_base;
-
- irq_base = IIC_NODE_STRIDE * spu->node + IIC_SPE_OFFSET;
-
- free_irq(irq_base + spu->isrc, spu);
- free_irq(irq_base + IIC_CLASS_STRIDE + spu->isrc, spu);
- free_irq(irq_base + 2*IIC_CLASS_STRIDE + spu->isrc, spu);
+ if (spu->irqs[0] != NO_IRQ)
+ free_irq(spu->irqs[0], spu);
+ if (spu->irqs[1] != NO_IRQ)
+ free_irq(spu->irqs[1], spu);
+ if (spu->irqs[2] != NO_IRQ)
+ free_irq(spu->irqs[2], spu);
}
-static LIST_HEAD(spu_list);
-static DECLARE_MUTEX(spu_mutex);
-
static void spu_init_channels(struct spu *spu)
{
static const struct {
{ 0x17, 1, }, { 0x18, 0, }, { 0x19, 0, }, { 0x1b, 0, },
{ 0x1c, 1, }, { 0x1d, 0, }, { 0x1e, 1, },
};
- struct spu_priv2 *priv2;
+ struct spu_priv2 __iomem *priv2;
int i;
priv2 = spu->priv2;
}
}
-static void spu_init_regs(struct spu *spu)
+struct spu *spu_alloc_node(int node)
{
- out_be64(&spu->priv1->int_mask_class0_RW, 0x7);
- out_be64(&spu->priv1->int_mask_class1_RW, 0x3);
- out_be64(&spu->priv1->int_mask_class2_RW, 0xe);
-}
+ struct spu *spu = NULL;
-struct spu *spu_alloc(void)
-{
- struct spu *spu;
-
- down(&spu_mutex);
- if (!list_empty(&spu_list)) {
- spu = list_entry(spu_list.next, struct spu, list);
+ mutex_lock(&spu_mutex);
+ if (!list_empty(&spu_list[node])) {
+ spu = list_entry(spu_list[node].next, struct spu, list);
list_del_init(&spu->list);
- pr_debug("Got SPU %x %d\n", spu->isrc, spu->number);
- } else {
- pr_debug("No SPU left\n");
- spu = NULL;
+ pr_debug("Got SPU %d %d\n", spu->number, spu->node);
}
- up(&spu_mutex);
+ mutex_unlock(&spu_mutex);
- if (spu) {
+ if (spu)
spu_init_channels(spu);
- spu_init_regs(spu);
- }
-
return spu;
}
-EXPORT_SYMBOL(spu_alloc);
-
-void spu_free(struct spu *spu)
-{
- down(&spu_mutex);
- spu->ibox_fasync = NULL;
- spu->wbox_fasync = NULL;
- list_add_tail(&spu->list, &spu_list);
- up(&spu_mutex);
-}
-EXPORT_SYMBOL(spu_free);
+EXPORT_SYMBOL_GPL(spu_alloc_node);
-extern int hash_page(unsigned long ea, unsigned long access, unsigned long trap); //XXX
-static int spu_handle_mm_fault(struct spu *spu)
+struct spu *spu_alloc(void)
{
- struct spu_priv1 __iomem *priv1;
- struct mm_struct *mm = spu->mm;
- struct vm_area_struct *vma;
- u64 ea, dsisr, is_write;
- int ret;
+ struct spu *spu = NULL;
+ int node;
- priv1 = spu->priv1;
- ea = in_be64(&priv1->mfc_dar_RW);
- dsisr = in_be64(&priv1->mfc_dsisr_RW);
-#if 0
- if (!IS_VALID_EA(ea)) {
- return -EFAULT;
- }
-#endif /* XXX */
- if (mm == NULL) {
- return -EFAULT;
- }
- if (mm->pgd == NULL) {
- return -EFAULT;
- }
-
- down_read(&mm->mmap_sem);
- vma = find_vma(mm, ea);
- if (!vma)
- goto bad_area;
- if (vma->vm_start <= ea)
- goto good_area;
- if (!(vma->vm_flags & VM_GROWSDOWN))
- goto bad_area;
-#if 0
- if (expand_stack(vma, ea))
- goto bad_area;
-#endif /* XXX */
-good_area:
- is_write = dsisr & MFC_DSISR_ACCESS_PUT;
- if (is_write) {
- if (!(vma->vm_flags & VM_WRITE))
- goto bad_area;
- } else {
- if (dsisr & MFC_DSISR_ACCESS_DENIED)
- goto bad_area;
- if (!(vma->vm_flags & (VM_READ | VM_EXEC)))
- goto bad_area;
- }
- ret = 0;
- switch (handle_mm_fault(mm, vma, ea, is_write)) {
- case VM_FAULT_MINOR:
- current->min_flt++;
- break;
- case VM_FAULT_MAJOR:
- current->maj_flt++;
- break;
- case VM_FAULT_SIGBUS:
- ret = -EFAULT;
- goto bad_area;
- case VM_FAULT_OOM:
- ret = -ENOMEM;
- goto bad_area;
- default:
- BUG();
+ for (node = 0; node < MAX_NUMNODES; node++) {
+ spu = spu_alloc_node(node);
+ if (spu)
+ break;
}
- up_read(&mm->mmap_sem);
- return ret;
-bad_area:
- up_read(&mm->mmap_sem);
- return -EFAULT;
+ return spu;
}
-static int spu_handle_pte_fault(struct spu *spu)
+void spu_free(struct spu *spu)
{
- struct spu_priv1 __iomem *priv1;
- u64 ea, dsisr, access, error = 0UL;
- int ret = 0;
-
- priv1 = spu->priv1;
- ea = in_be64(&priv1->mfc_dar_RW);
- dsisr = in_be64(&priv1->mfc_dsisr_RW);
- access = (_PAGE_PRESENT | _PAGE_USER);
- if (dsisr & MFC_DSISR_PTE_NOT_FOUND) {
- if (hash_page(ea, access, 0x300) != 0)
- error |= CLASS1_ENABLE_STORAGE_FAULT_INTR;
- }
- if ((error & CLASS1_ENABLE_STORAGE_FAULT_INTR) ||
- (dsisr & MFC_DSISR_ACCESS_DENIED)) {
- if ((ret = spu_handle_mm_fault(spu)) != 0)
- error |= CLASS1_ENABLE_STORAGE_FAULT_INTR;
- else
- error &= ~CLASS1_ENABLE_STORAGE_FAULT_INTR;
- }
- if (!error)
- spu_restart_dma(spu);
-
- return ret;
+ mutex_lock(&spu_mutex);
+ list_add_tail(&spu->list, &spu_list[spu->node]);
+ mutex_unlock(&spu_mutex);
}
+EXPORT_SYMBOL_GPL(spu_free);
-int spu_run(struct spu *spu)
+static int spu_shutdown(struct sys_device *sysdev)
{
- struct spu_problem __iomem *prob;
- struct spu_priv1 __iomem *priv1;
- struct spu_priv2 __iomem *priv2;
- unsigned long status;
- int ret;
-
- prob = spu->problem;
- priv1 = spu->priv1;
- priv2 = spu->priv2;
+ struct spu *spu = container_of(sysdev, struct spu, sysdev);
- /* Let SPU run. */
- spu->mm = current->mm;
- eieio();
- out_be32(&prob->spu_runcntl_RW, SPU_RUNCNTL_RUNNABLE);
-
- do {
- ret = wait_event_interruptible(spu->stop_wq,
- (!((status = in_be32(&prob->spu_status_R)) & 0x1))
- || (in_be64(&priv1->mfc_dsisr_RW) & MFC_DSISR_PTE_NOT_FOUND)
- || spu->class_0_pending);
-
- if (status & SPU_STATUS_STOPPED_BY_STOP)
- ret = -EAGAIN;
- else if (status & SPU_STATUS_STOPPED_BY_HALT)
- ret = -EIO;
- else if (in_be64(&priv1->mfc_dsisr_RW) & MFC_DSISR_PTE_NOT_FOUND)
- ret = spu_handle_pte_fault(spu);
-
- if (spu->class_0_pending)
- spu_irq_class_0_bottom(spu);
-
- if (!ret && signal_pending(current))
- ret = -ERESTARTSYS;
-
- } while (!ret);
-
- /* Ensure SPU is stopped. */
- out_be32(&prob->spu_runcntl_RW, SPU_RUNCNTL_STOP);
- eieio();
- while (in_be32(&prob->spu_status_R) & SPU_STATUS_RUNNING)
- cpu_relax();
-
- out_be64(&priv2->slb_invalidate_all_W, 0);
- out_be64(&priv1->tlb_invalidate_entry_W, 0UL);
- eieio();
-
- spu->mm = NULL;
-
- /* Check for SPU breakpoint. */
- if (unlikely(current->ptrace & PT_PTRACED)) {
- status = in_be32(&prob->spu_status_R);
-
- if ((status & SPU_STATUS_STOPPED_BY_STOP)
- && status >> SPU_STOP_STATUS_SHIFT == 0x3fff) {
- force_sig(SIGTRAP, current);
- ret = -ERESTARTSYS;
- }
- }
-
- return ret;
+ spu_free_irqs(spu);
+ spu_destroy_spu(spu);
+ return 0;
}
-EXPORT_SYMBOL(spu_run);
-
-static void __iomem * __init map_spe_prop(struct device_node *n,
- const char *name)
-{
- struct address_prop {
- unsigned long address;
- unsigned int len;
- } __attribute__((packed)) *prop;
- void *p;
- int proplen;
+struct sysdev_class spu_sysdev_class = {
+ set_kset_name("spu"),
+ .shutdown = spu_shutdown,
+};
- p = get_property(n, name, &proplen);
- if (proplen != sizeof (struct address_prop))
- return NULL;
+int spu_add_sysdev_attr(struct sysdev_attribute *attr)
+{
+ struct spu *spu;
+ mutex_lock(&spu_mutex);
- prop = p;
+ list_for_each_entry(spu, &spu_full_list, full_list)
+ sysdev_create_file(&spu->sysdev, attr);
- return ioremap(prop->address, prop->len);
+ mutex_unlock(&spu_mutex);
+ return 0;
}
+EXPORT_SYMBOL_GPL(spu_add_sysdev_attr);
-static void spu_unmap(struct spu *spu)
+int spu_add_sysdev_attr_group(struct attribute_group *attrs)
{
- iounmap(spu->priv2);
- iounmap(spu->priv1);
- iounmap(spu->problem);
- iounmap((u8 __iomem *)spu->local_store);
-}
+ struct spu *spu;
+ mutex_lock(&spu_mutex);
-static int __init spu_map_device(struct spu *spu, struct device_node *spe)
-{
- char *prop;
- int ret;
+ list_for_each_entry(spu, &spu_full_list, full_list)
+ sysfs_create_group(&spu->sysdev.kobj, attrs);
- ret = -ENODEV;
- prop = get_property(spe, "isrc", NULL);
- if (!prop)
- goto out;
- spu->isrc = *(unsigned int *)prop;
+ mutex_unlock(&spu_mutex);
+ return 0;
+}
+EXPORT_SYMBOL_GPL(spu_add_sysdev_attr_group);
- spu->name = get_property(spe, "name", NULL);
- if (!spu->name)
- goto out;
- prop = get_property(spe, "local-store", NULL);
- if (!prop)
- goto out;
- spu->local_store_phys = *(unsigned long *)prop;
+void spu_remove_sysdev_attr(struct sysdev_attribute *attr)
+{
+ struct spu *spu;
+ mutex_lock(&spu_mutex);
- /* we use local store as ram, not io memory */
- spu->local_store = (void __force *)map_spe_prop(spe, "local-store");
- if (!spu->local_store)
- goto out;
+ list_for_each_entry(spu, &spu_full_list, full_list)
+ sysdev_remove_file(&spu->sysdev, attr);
- spu->problem= map_spe_prop(spe, "problem");
- if (!spu->problem)
- goto out_unmap;
+ mutex_unlock(&spu_mutex);
+}
+EXPORT_SYMBOL_GPL(spu_remove_sysdev_attr);
- spu->priv1= map_spe_prop(spe, "priv1");
- if (!spu->priv1)
- goto out_unmap;
+void spu_remove_sysdev_attr_group(struct attribute_group *attrs)
+{
+ struct spu *spu;
+ mutex_lock(&spu_mutex);
- spu->priv2= map_spe_prop(spe, "priv2");
- if (!spu->priv2)
- goto out_unmap;
- ret = 0;
- goto out;
+ list_for_each_entry(spu, &spu_full_list, full_list)
+ sysfs_remove_group(&spu->sysdev.kobj, attrs);
-out_unmap:
- spu_unmap(spu);
-out:
- return ret;
+ mutex_unlock(&spu_mutex);
}
+EXPORT_SYMBOL_GPL(spu_remove_sysdev_attr_group);
-static int __init find_spu_node_id(struct device_node *spe)
+static int spu_create_sysdev(struct spu *spu)
{
- unsigned int *id;
- struct device_node *cpu;
+ int ret;
- cpu = spe->parent->parent;
- id = (unsigned int *)get_property(cpu, "node-id", NULL);
+ spu->sysdev.id = spu->number;
+ spu->sysdev.cls = &spu_sysdev_class;
+ ret = sysdev_register(&spu->sysdev);
+ if (ret) {
+ printk(KERN_ERR "Can't register SPU %d with sysfs\n",
+ spu->number);
+ return ret;
+ }
- return id ? *id : 0;
+ sysfs_add_device_to_node(&spu->sysdev, spu->node);
+
+ return 0;
}
-static int __init create_spu(struct device_node *spe)
+static int __init create_spu(void *data)
{
struct spu *spu;
int ret;
static int number;
+ unsigned long flags;
ret = -ENOMEM;
- spu = kmalloc(sizeof (*spu), GFP_KERNEL);
+ spu = kzalloc(sizeof (*spu), GFP_KERNEL);
if (!spu)
goto out;
- ret = spu_map_device(spu, spe);
- if (ret)
- goto out_free;
-
- spu->node = find_spu_node_id(spe);
- spu->stop_code = 0;
- spu->slb_replace = 0;
- spu->mm = NULL;
- spu->class_0_pending = 0;
spin_lock_init(&spu->register_lock);
+ mutex_lock(&spu_mutex);
+ spu->number = number++;
+ mutex_unlock(&spu_mutex);
- out_be64(&spu->priv1->mfc_sdr_RW, mfspr(SPRN_SDR1));
- out_be64(&spu->priv1->mfc_sr1_RW, 0x33);
-
- init_waitqueue_head(&spu->stop_wq);
- init_waitqueue_head(&spu->wbox_wq);
- init_waitqueue_head(&spu->ibox_wq);
+ ret = spu_create_spu(spu, data);
- spu->ibox_fasync = NULL;
- spu->wbox_fasync = NULL;
+ if (ret)
+ goto out_free;
- down(&spu_mutex);
- spu->number = number++;
+ spu_mfc_sdr_setup(spu);
+ spu_mfc_sr1_set(spu, 0x33);
ret = spu_request_irqs(spu);
if (ret)
- goto out_unmap;
+ goto out_destroy;
+
+ ret = spu_create_sysdev(spu);
+ if (ret)
+ goto out_free_irqs;
- list_add(&spu->list, &spu_list);
- up(&spu_mutex);
+ mutex_lock(&spu_mutex);
+ spin_lock_irqsave(&spu_list_lock, flags);
+ list_add(&spu->list, &spu_list[spu->node]);
+ list_add(&spu->full_list, &spu_full_list);
+ spin_unlock_irqrestore(&spu_list_lock, flags);
+ mutex_unlock(&spu_mutex);
+
+ spu->stats.utilization_state = SPU_UTIL_IDLE;
+ spu->stats.tstamp = jiffies;
- pr_debug(KERN_DEBUG "Using SPE %s %02x %p %p %p %p %d\n",
- spu->name, spu->isrc, spu->local_store,
- spu->problem, spu->priv1, spu->priv2, spu->number);
goto out;
-out_unmap:
- up(&spu_mutex);
- spu_unmap(spu);
+out_free_irqs:
+ spu_free_irqs(spu);
+out_destroy:
+ spu_destroy_spu(spu);
out_free:
kfree(spu);
out:
return ret;
}
-static void destroy_spu(struct spu *spu)
+static const char *spu_state_names[] = {
+ "user", "system", "iowait", "idle"
+};
+
+static unsigned long long spu_acct_time(struct spu *spu,
+ enum spu_utilization_state state)
{
- list_del_init(&spu->list);
+ unsigned long long time = spu->stats.times[state];
- spu_free_irqs(spu);
- spu_unmap(spu);
- kfree(spu);
+ if (spu->stats.utilization_state == state)
+ time += jiffies - spu->stats.tstamp;
+
+ return jiffies_to_msecs(time);
}
-static void cleanup_spu_base(void)
+
+static ssize_t spu_stat_show(struct sys_device *sysdev, char *buf)
{
- struct spu *spu, *tmp;
- down(&spu_mutex);
- list_for_each_entry_safe(spu, tmp, &spu_list, list)
- destroy_spu(spu);
- up(&spu_mutex);
+ struct spu *spu = container_of(sysdev, struct spu, sysdev);
+
+ return sprintf(buf, "%s %llu %llu %llu %llu "
+ "%llu %llu %llu %llu %llu %llu %llu %llu\n",
+ spu_state_names[spu->stats.utilization_state],
+ spu_acct_time(spu, SPU_UTIL_USER),
+ spu_acct_time(spu, SPU_UTIL_SYSTEM),
+ spu_acct_time(spu, SPU_UTIL_IOWAIT),
+ spu_acct_time(spu, SPU_UTIL_IDLE),
+ spu->stats.vol_ctx_switch,
+ spu->stats.invol_ctx_switch,
+ spu->stats.slb_flt,
+ spu->stats.hash_flt,
+ spu->stats.min_flt,
+ spu->stats.maj_flt,
+ spu->stats.class2_intr,
+ spu->stats.libassist);
}
-module_exit(cleanup_spu_base);
+
+static SYSDEV_ATTR(stat, 0644, spu_stat_show, NULL);
static int __init init_spu_base(void)
{
- struct device_node *node;
- int ret;
+ int i, ret = 0;
- ret = -ENODEV;
- for (node = of_find_node_by_type(NULL, "spe");
- node; node = of_find_node_by_type(node, "spe")) {
- ret = create_spu(node);
- if (ret) {
- printk(KERN_WARNING "%s: Error initializing %s\n",
- __FUNCTION__, node->name);
- cleanup_spu_base();
- break;
- }
+ for (i = 0; i < MAX_NUMNODES; i++)
+ INIT_LIST_HEAD(&spu_list[i]);
+
+ if (!spu_management_ops)
+ goto out;
+
+ /* create sysdev class for spus */
+ ret = sysdev_class_register(&spu_sysdev_class);
+ if (ret)
+ goto out;
+
+ ret = spu_enumerate_spus(create_spu);
+
+ if (ret < 0) {
+ printk(KERN_WARNING "%s: Error initializing spus\n",
+ __FUNCTION__);
+ goto out_unregister_sysdev_class;
}
- /* in some old firmware versions, the spe is called 'spc', so we
- look for that as well */
- for (node = of_find_node_by_type(NULL, "spc");
- node; node = of_find_node_by_type(node, "spc")) {
- ret = create_spu(node);
- if (ret) {
- printk(KERN_WARNING "%s: Error initializing %s\n",
- __FUNCTION__, node->name);
- cleanup_spu_base();
- break;
- }
+
+ if (ret > 0) {
+ /*
+ * We cannot put the forward declaration in
+ * <linux/linux_logo.h> because of conflicting session type
+ * conflicts for const and __initdata with different compiler
+ * versions
+ */
+ extern const struct linux_logo logo_spe_clut224;
+
+ fb_append_extra_logo(&logo_spe_clut224, ret);
}
+
+ xmon_register_spus(&spu_full_list);
+ crash_register_spus(&spu_full_list);
+ spu_add_sysdev_attr(&attr_stat);
+
+ return 0;
+
+ out_unregister_sysdev_class:
+ sysdev_class_unregister(&spu_sysdev_class);
+ out:
+
return ret;
}
module_init(init_spu_base);
git://ftp.safe.ca / safe / jmp / linux-2.6 / blobdiff