Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dtor/input
[pandora-kernel.git] / arch / sparc64 / kernel / pci_sun4v.c
index 639cf06..8c4875b 100644 (file)
@@ -33,30 +33,30 @@ static unsigned long vpci_minor = 1;
 #define PGLIST_NENTS   (PAGE_SIZE / sizeof(u64))
 
 struct iommu_batch {
-       struct pci_dev  *pdev;          /* Device mapping is for.       */
+       struct device   *dev;           /* Device mapping is for.       */
        unsigned long   prot;           /* IOMMU page protections       */
        unsigned long   entry;          /* Index into IOTSB.            */
        u64             *pglist;        /* List of physical pages       */
        unsigned long   npages;         /* Number of pages in list.     */
 };
 
-static DEFINE_PER_CPU(struct iommu_batch, pci_iommu_batch);
+static DEFINE_PER_CPU(struct iommu_batch, iommu_batch);
 
 /* Interrupts must be disabled.  */
-static inline void pci_iommu_batch_start(struct pci_dev *pdev, unsigned long prot, unsigned long entry)
+static inline void iommu_batch_start(struct device *dev, unsigned long prot, unsigned long entry)
 {
-       struct iommu_batch *p = &__get_cpu_var(pci_iommu_batch);
+       struct iommu_batch *p = &__get_cpu_var(iommu_batch);
 
-       p->pdev         = pdev;
+       p->dev          = dev;
        p->prot         = prot;
        p->entry        = entry;
        p->npages       = 0;
 }
 
 /* Interrupts must be disabled.  */
-static long pci_iommu_batch_flush(struct iommu_batch *p)
+static long iommu_batch_flush(struct iommu_batch *p)
 {
-       struct pci_pbm_info *pbm = p->pdev->dev.archdata.host_controller;
+       struct pci_pbm_info *pbm = p->dev->archdata.host_controller;
        unsigned long devhandle = pbm->devhandle;
        unsigned long prot = p->prot;
        unsigned long entry = p->entry;
@@ -70,7 +70,7 @@ static long pci_iommu_batch_flush(struct iommu_batch *p)
                                          npages, prot, __pa(pglist));
                if (unlikely(num < 0)) {
                        if (printk_ratelimit())
-                               printk("pci_iommu_batch_flush: IOMMU map of "
+                               printk("iommu_batch_flush: IOMMU map of "
                                       "[%08lx:%08lx:%lx:%lx:%lx] failed with "
                                       "status %ld\n",
                                       devhandle, HV_PCI_TSBID(0, entry),
@@ -90,30 +90,30 @@ static long pci_iommu_batch_flush(struct iommu_batch *p)
 }
 
 /* Interrupts must be disabled.  */
-static inline long pci_iommu_batch_add(u64 phys_page)
+static inline long iommu_batch_add(u64 phys_page)
 {
-       struct iommu_batch *p = &__get_cpu_var(pci_iommu_batch);
+       struct iommu_batch *p = &__get_cpu_var(iommu_batch);
 
        BUG_ON(p->npages >= PGLIST_NENTS);
 
        p->pglist[p->npages++] = phys_page;
        if (p->npages == PGLIST_NENTS)
-               return pci_iommu_batch_flush(p);
+               return iommu_batch_flush(p);
 
        return 0;
 }
 
 /* Interrupts must be disabled.  */
-static inline long pci_iommu_batch_end(void)
+static inline long iommu_batch_end(void)
 {
-       struct iommu_batch *p = &__get_cpu_var(pci_iommu_batch);
+       struct iommu_batch *p = &__get_cpu_var(iommu_batch);
 
        BUG_ON(p->npages >= PGLIST_NENTS);
 
-       return pci_iommu_batch_flush(p);
+       return iommu_batch_flush(p);
 }
 
-static long pci_arena_alloc(struct iommu_arena *arena, unsigned long npages)
+static long arena_alloc(struct iommu_arena *arena, unsigned long npages)
 {
        unsigned long n, i, start, end, limit;
        int pass;
@@ -152,7 +152,8 @@ again:
        return n;
 }
 
-static void pci_arena_free(struct iommu_arena *arena, unsigned long base, unsigned long npages)
+static void arena_free(struct iommu_arena *arena, unsigned long base,
+                      unsigned long npages)
 {
        unsigned long i;
 
@@ -160,7 +161,8 @@ static void pci_arena_free(struct iommu_arena *arena, unsigned long base, unsign
                __clear_bit(i, arena->map);
 }
 
-static void *pci_4v_alloc_consistent(struct pci_dev *pdev, size_t size, dma_addr_t *dma_addrp, gfp_t gfp)
+static void *dma_4v_alloc_coherent(struct device *dev, size_t size,
+                                  dma_addr_t *dma_addrp, gfp_t gfp)
 {
        struct iommu *iommu;
        unsigned long flags, order, first_page, npages, n;
@@ -180,10 +182,10 @@ static void *pci_4v_alloc_consistent(struct pci_dev *pdev, size_t size, dma_addr
 
        memset((char *)first_page, 0, PAGE_SIZE << order);
 
-       iommu = pdev->dev.archdata.iommu;
+       iommu = dev->archdata.iommu;
 
        spin_lock_irqsave(&iommu->lock, flags);
-       entry = pci_arena_alloc(&iommu->arena, npages);
+       entry = arena_alloc(&iommu->arena, npages);
        spin_unlock_irqrestore(&iommu->lock, flags);
 
        if (unlikely(entry < 0L))
@@ -196,18 +198,18 @@ static void *pci_4v_alloc_consistent(struct pci_dev *pdev, size_t size, dma_addr
 
        local_irq_save(flags);
 
-       pci_iommu_batch_start(pdev,
-                             (HV_PCI_MAP_ATTR_READ |
-                              HV_PCI_MAP_ATTR_WRITE),
-                             entry);
+       iommu_batch_start(dev,
+                         (HV_PCI_MAP_ATTR_READ |
+                          HV_PCI_MAP_ATTR_WRITE),
+                         entry);
 
        for (n = 0; n < npages; n++) {
-               long err = pci_iommu_batch_add(first_page + (n * PAGE_SIZE));
+               long err = iommu_batch_add(first_page + (n * PAGE_SIZE));
                if (unlikely(err < 0L))
                        goto iommu_map_fail;
        }
 
-       if (unlikely(pci_iommu_batch_end() < 0L))
+       if (unlikely(iommu_batch_end() < 0L))
                goto iommu_map_fail;
 
        local_irq_restore(flags);
@@ -217,7 +219,7 @@ static void *pci_4v_alloc_consistent(struct pci_dev *pdev, size_t size, dma_addr
 iommu_map_fail:
        /* Interrupts are disabled.  */
        spin_lock(&iommu->lock);
-       pci_arena_free(&iommu->arena, entry, npages);
+       arena_free(&iommu->arena, entry, npages);
        spin_unlock_irqrestore(&iommu->lock, flags);
 
 arena_alloc_fail:
@@ -225,7 +227,8 @@ arena_alloc_fail:
        return NULL;
 }
 
-static void pci_4v_free_consistent(struct pci_dev *pdev, size_t size, void *cpu, dma_addr_t dvma)
+static void dma_4v_free_coherent(struct device *dev, size_t size, void *cpu,
+                                dma_addr_t dvma)
 {
        struct pci_pbm_info *pbm;
        struct iommu *iommu;
@@ -233,14 +236,14 @@ static void pci_4v_free_consistent(struct pci_dev *pdev, size_t size, void *cpu,
        u32 devhandle;
 
        npages = IO_PAGE_ALIGN(size) >> IO_PAGE_SHIFT;
-       iommu = pdev->dev.archdata.iommu;
-       pbm = pdev->dev.archdata.host_controller;
+       iommu = dev->archdata.iommu;
+       pbm = dev->archdata.host_controller;
        devhandle = pbm->devhandle;
        entry = ((dvma - iommu->page_table_map_base) >> IO_PAGE_SHIFT);
 
        spin_lock_irqsave(&iommu->lock, flags);
 
-       pci_arena_free(&iommu->arena, entry, npages);
+       arena_free(&iommu->arena, entry, npages);
 
        do {
                unsigned long num;
@@ -258,7 +261,8 @@ static void pci_4v_free_consistent(struct pci_dev *pdev, size_t size, void *cpu,
                free_pages((unsigned long)cpu, order);
 }
 
-static dma_addr_t pci_4v_map_single(struct pci_dev *pdev, void *ptr, size_t sz, int direction)
+static dma_addr_t dma_4v_map_single(struct device *dev, void *ptr, size_t sz,
+                                   enum dma_data_direction direction)
 {
        struct iommu *iommu;
        unsigned long flags, npages, oaddr;
@@ -267,9 +271,9 @@ static dma_addr_t pci_4v_map_single(struct pci_dev *pdev, void *ptr, size_t sz,
        unsigned long prot;
        long entry;
 
-       iommu = pdev->dev.archdata.iommu;
+       iommu = dev->archdata.iommu;
 
-       if (unlikely(direction == PCI_DMA_NONE))
+       if (unlikely(direction == DMA_NONE))
                goto bad;
 
        oaddr = (unsigned long)ptr;
@@ -277,7 +281,7 @@ static dma_addr_t pci_4v_map_single(struct pci_dev *pdev, void *ptr, size_t sz,
        npages >>= IO_PAGE_SHIFT;
 
        spin_lock_irqsave(&iommu->lock, flags);
-       entry = pci_arena_alloc(&iommu->arena, npages);
+       entry = arena_alloc(&iommu->arena, npages);
        spin_unlock_irqrestore(&iommu->lock, flags);
 
        if (unlikely(entry < 0L))
@@ -288,19 +292,19 @@ static dma_addr_t pci_4v_map_single(struct pci_dev *pdev, void *ptr, size_t sz,
        ret = bus_addr | (oaddr & ~IO_PAGE_MASK);
        base_paddr = __pa(oaddr & IO_PAGE_MASK);
        prot = HV_PCI_MAP_ATTR_READ;
-       if (direction != PCI_DMA_TODEVICE)
+       if (direction != DMA_TO_DEVICE)
                prot |= HV_PCI_MAP_ATTR_WRITE;
 
        local_irq_save(flags);
 
-       pci_iommu_batch_start(pdev, prot, entry);
+       iommu_batch_start(dev, prot, entry);
 
        for (i = 0; i < npages; i++, base_paddr += IO_PAGE_SIZE) {
-               long err = pci_iommu_batch_add(base_paddr);
+               long err = iommu_batch_add(base_paddr);
                if (unlikely(err < 0L))
                        goto iommu_map_fail;
        }
-       if (unlikely(pci_iommu_batch_end() < 0L))
+       if (unlikely(iommu_batch_end() < 0L))
                goto iommu_map_fail;
 
        local_irq_restore(flags);
@@ -310,18 +314,19 @@ static dma_addr_t pci_4v_map_single(struct pci_dev *pdev, void *ptr, size_t sz,
 bad:
        if (printk_ratelimit())
                WARN_ON(1);
-       return PCI_DMA_ERROR_CODE;
+       return DMA_ERROR_CODE;
 
 iommu_map_fail:
        /* Interrupts are disabled.  */
        spin_lock(&iommu->lock);
-       pci_arena_free(&iommu->arena, entry, npages);
+       arena_free(&iommu->arena, entry, npages);
        spin_unlock_irqrestore(&iommu->lock, flags);
 
-       return PCI_DMA_ERROR_CODE;
+       return DMA_ERROR_CODE;
 }
 
-static void pci_4v_unmap_single(struct pci_dev *pdev, dma_addr_t bus_addr, size_t sz, int direction)
+static void dma_4v_unmap_single(struct device *dev, dma_addr_t bus_addr,
+                               size_t sz, enum dma_data_direction direction)
 {
        struct pci_pbm_info *pbm;
        struct iommu *iommu;
@@ -329,14 +334,14 @@ static void pci_4v_unmap_single(struct pci_dev *pdev, dma_addr_t bus_addr, size_
        long entry;
        u32 devhandle;
 
-       if (unlikely(direction == PCI_DMA_NONE)) {
+       if (unlikely(direction == DMA_NONE)) {
                if (printk_ratelimit())
                        WARN_ON(1);
                return;
        }
 
-       iommu = pdev->dev.archdata.iommu;
-       pbm = pdev->dev.archdata.host_controller;
+       iommu = dev->archdata.iommu;
+       pbm = dev->archdata.host_controller;
        devhandle = pbm->devhandle;
 
        npages = IO_PAGE_ALIGN(bus_addr + sz) - (bus_addr & IO_PAGE_MASK);
@@ -346,7 +351,7 @@ static void pci_4v_unmap_single(struct pci_dev *pdev, dma_addr_t bus_addr, size_
        spin_lock_irqsave(&iommu->lock, flags);
 
        entry = (bus_addr - iommu->page_table_map_base) >> IO_PAGE_SHIFT;
-       pci_arena_free(&iommu->arena, entry, npages);
+       arena_free(&iommu->arena, entry, npages);
 
        do {
                unsigned long num;
@@ -360,21 +365,19 @@ static void pci_4v_unmap_single(struct pci_dev *pdev, dma_addr_t bus_addr, size_
        spin_unlock_irqrestore(&iommu->lock, flags);
 }
 
-#define SG_ENT_PHYS_ADDRESS(SG)        \
-       (__pa(page_address((SG)->page)) + (SG)->offset)
+#define SG_ENT_PHYS_ADDRESS(SG)        (__pa(sg_virt((SG))))
 
-static inline long fill_sg(long entry, struct pci_dev *pdev,
-                          struct scatterlist *sg,
-                          int nused, int nelems, unsigned long prot)
+static long fill_sg(long entry, struct device *dev,
+                   struct scatterlist *sg,
+                   int nused, int nelems, unsigned long prot)
 {
        struct scatterlist *dma_sg = sg;
-       struct scatterlist *sg_end = sg + nelems;
        unsigned long flags;
        int i;
 
        local_irq_save(flags);
 
-       pci_iommu_batch_start(pdev, prot, entry);
+       iommu_batch_start(dev, prot, entry);
 
        for (i = 0; i < nused; i++) {
                unsigned long pteval = ~0UL;
@@ -408,14 +411,15 @@ static inline long fill_sg(long entry, struct pci_dev *pdev,
                                        len -= (IO_PAGE_SIZE - (tmp & (IO_PAGE_SIZE - 1UL)));
                                        break;
                                }
-                               sg++;
+                               sg = sg_next(sg);
+                               nelems--;
                        }
 
                        pteval = (pteval & IOPTE_PAGE);
                        while (len > 0) {
                                long err;
 
-                               err = pci_iommu_batch_add(pteval);
+                               err = iommu_batch_add(pteval);
                                if (unlikely(err < 0L))
                                        goto iommu_map_failed;
 
@@ -426,27 +430,29 @@ static inline long fill_sg(long entry, struct pci_dev *pdev,
                        }
 
                        pteval = (pteval & IOPTE_PAGE) + len;
-                       sg++;
+                       sg = sg_next(sg);
+                       nelems--;
 
                        /* Skip over any tail mappings we've fully mapped,
                         * adjusting pteval along the way.  Stop when we
                         * detect a page crossing event.
                         */
-                       while (sg < sg_end &&
+                       while (nelems &&
                               (pteval << (64 - IO_PAGE_SHIFT)) != 0UL &&
                               (pteval == SG_ENT_PHYS_ADDRESS(sg)) &&
                               ((pteval ^
                                 (SG_ENT_PHYS_ADDRESS(sg) + sg->length - 1UL)) >> IO_PAGE_SHIFT) == 0UL) {
                                pteval += sg->length;
-                               sg++;
+                               sg = sg_next(sg);
+                               nelems--;
                        }
                        if ((pteval << (64 - IO_PAGE_SHIFT)) == 0UL)
                                pteval = ~0UL;
                } while (dma_npages != 0);
-               dma_sg++;
+               dma_sg = sg_next(dma_sg);
        }
 
-       if (unlikely(pci_iommu_batch_end() < 0L))
+       if (unlikely(iommu_batch_end() < 0L))
                goto iommu_map_failed;
 
        local_irq_restore(flags);
@@ -457,7 +463,8 @@ iommu_map_failed:
        return -1L;
 }
 
-static int pci_4v_map_sg(struct pci_dev *pdev, struct scatterlist *sglist, int nelems, int direction)
+static int dma_4v_map_sg(struct device *dev, struct scatterlist *sglist,
+                        int nelems, enum dma_data_direction direction)
 {
        struct iommu *iommu;
        unsigned long flags, npages, prot;
@@ -469,18 +476,17 @@ static int pci_4v_map_sg(struct pci_dev *pdev, struct scatterlist *sglist, int n
        /* Fast path single entry scatterlists. */
        if (nelems == 1) {
                sglist->dma_address =
-                       pci_4v_map_single(pdev,
-                                         (page_address(sglist->page) + sglist->offset),
+                       dma_4v_map_single(dev, sg_virt(sglist),
                                          sglist->length, direction);
-               if (unlikely(sglist->dma_address == PCI_DMA_ERROR_CODE))
+               if (unlikely(sglist->dma_address == DMA_ERROR_CODE))
                        return 0;
                sglist->dma_length = sglist->length;
                return 1;
        }
 
-       iommu = pdev->dev.archdata.iommu;
+       iommu = dev->archdata.iommu;
        
-       if (unlikely(direction == PCI_DMA_NONE))
+       if (unlikely(direction == DMA_NONE))
                goto bad;
 
        /* Step 1: Prepare scatter list. */
@@ -488,7 +494,7 @@ static int pci_4v_map_sg(struct pci_dev *pdev, struct scatterlist *sglist, int n
 
        /* Step 2: Allocate a cluster and context, if necessary. */
        spin_lock_irqsave(&iommu->lock, flags);
-       entry = pci_arena_alloc(&iommu->arena, npages);
+       entry = arena_alloc(&iommu->arena, npages);
        spin_unlock_irqrestore(&iommu->lock, flags);
 
        if (unlikely(entry < 0L))
@@ -503,17 +509,17 @@ static int pci_4v_map_sg(struct pci_dev *pdev, struct scatterlist *sglist, int n
        sgtmp = sglist;
        while (used && sgtmp->dma_length) {
                sgtmp->dma_address += dma_base;
-               sgtmp++;
+               sgtmp = sg_next(sgtmp);
                used--;
        }
        used = nelems - used;
 
        /* Step 4: Create the mappings. */
        prot = HV_PCI_MAP_ATTR_READ;
-       if (direction != PCI_DMA_TODEVICE)
+       if (direction != DMA_TO_DEVICE)
                prot |= HV_PCI_MAP_ATTR_WRITE;
 
-       err = fill_sg(entry, pdev, sglist, used, nelems, prot);
+       err = fill_sg(entry, dev, sglist, used, nelems, prot);
        if (unlikely(err < 0L))
                goto iommu_map_failed;
 
@@ -526,43 +532,48 @@ bad:
 
 iommu_map_failed:
        spin_lock_irqsave(&iommu->lock, flags);
-       pci_arena_free(&iommu->arena, entry, npages);
+       arena_free(&iommu->arena, entry, npages);
        spin_unlock_irqrestore(&iommu->lock, flags);
 
        return 0;
 }
 
-static void pci_4v_unmap_sg(struct pci_dev *pdev, struct scatterlist *sglist, int nelems, int direction)
+static void dma_4v_unmap_sg(struct device *dev, struct scatterlist *sglist,
+                           int nelems, enum dma_data_direction direction)
 {
        struct pci_pbm_info *pbm;
        struct iommu *iommu;
        unsigned long flags, i, npages;
+       struct scatterlist *sg, *sgprv;
        long entry;
        u32 devhandle, bus_addr;
 
-       if (unlikely(direction == PCI_DMA_NONE)) {
+       if (unlikely(direction == DMA_NONE)) {
                if (printk_ratelimit())
                        WARN_ON(1);
        }
 
-       iommu = pdev->dev.archdata.iommu;
-       pbm = pdev->dev.archdata.host_controller;
+       iommu = dev->archdata.iommu;
+       pbm = dev->archdata.host_controller;
        devhandle = pbm->devhandle;
        
        bus_addr = sglist->dma_address & IO_PAGE_MASK;
-
-       for (i = 1; i < nelems; i++)
-               if (sglist[i].dma_length == 0)
+       sgprv = NULL;
+       for_each_sg(sglist, sg, nelems, i) {
+               if (sg->dma_length == 0)
                        break;
-       i--;
-       npages = (IO_PAGE_ALIGN(sglist[i].dma_address + sglist[i].dma_length) -
+
+               sgprv = sg;
+       }
+
+       npages = (IO_PAGE_ALIGN(sgprv->dma_address + sgprv->dma_length) -
                  bus_addr) >> IO_PAGE_SHIFT;
 
        entry = ((bus_addr - iommu->page_table_map_base) >> IO_PAGE_SHIFT);
 
        spin_lock_irqsave(&iommu->lock, flags);
 
-       pci_arena_free(&iommu->arena, entry, npages);
+       arena_free(&iommu->arena, entry, npages);
 
        do {
                unsigned long num;
@@ -576,25 +587,29 @@ static void pci_4v_unmap_sg(struct pci_dev *pdev, struct scatterlist *sglist, in
        spin_unlock_irqrestore(&iommu->lock, flags);
 }
 
-static void pci_4v_dma_sync_single_for_cpu(struct pci_dev *pdev, dma_addr_t bus_addr, size_t sz, int direction)
+static void dma_4v_sync_single_for_cpu(struct device *dev,
+                                      dma_addr_t bus_addr, size_t sz,
+                                      enum dma_data_direction direction)
 {
        /* Nothing to do... */
 }
 
-static void pci_4v_dma_sync_sg_for_cpu(struct pci_dev *pdev, struct scatterlist *sglist, int nelems, int direction)
+static void dma_4v_sync_sg_for_cpu(struct device *dev,
+                                  struct scatterlist *sglist, int nelems,
+                                  enum dma_data_direction direction)
 {
        /* Nothing to do... */
 }
 
-const struct pci_iommu_ops pci_sun4v_iommu_ops = {
-       .alloc_consistent               = pci_4v_alloc_consistent,
-       .free_consistent                = pci_4v_free_consistent,
-       .map_single                     = pci_4v_map_single,
-       .unmap_single                   = pci_4v_unmap_single,
-       .map_sg                         = pci_4v_map_sg,
-       .unmap_sg                       = pci_4v_unmap_sg,
-       .dma_sync_single_for_cpu        = pci_4v_dma_sync_single_for_cpu,
-       .dma_sync_sg_for_cpu            = pci_4v_dma_sync_sg_for_cpu,
+const struct dma_ops sun4v_dma_ops = {
+       .alloc_coherent                 = dma_4v_alloc_coherent,
+       .free_coherent                  = dma_4v_free_coherent,
+       .map_single                     = dma_4v_map_single,
+       .unmap_single                   = dma_4v_unmap_single,
+       .map_sg                         = dma_4v_map_sg,
+       .unmap_sg                       = dma_4v_unmap_sg,
+       .sync_single_for_cpu            = dma_4v_sync_single_for_cpu,
+       .sync_sg_for_cpu                = dma_4v_sync_sg_for_cpu,
 };
 
 static void pci_sun4v_scan_bus(struct pci_pbm_info *pbm)
@@ -736,111 +751,102 @@ struct pci_sun4v_msiq_entry {
        u64             reserved2;
 };
 
-/* For now this just runs as a pre-handler for the real interrupt handler.
- * So we just walk through the queue and ACK all the entries, update the
- * head pointer, and return.
- *
- * In the longer term it would be nice to do something more integrated
- * wherein we can pass in some of this MSI info to the drivers.  This
- * would be most useful for PCIe fabric error messages, although we could
- * invoke those directly from the loop here in order to pass the info around.
- */
-static void pci_sun4v_msi_prehandler(unsigned int ino, void *data1, void *data2)
+static int pci_sun4v_get_head(struct pci_pbm_info *pbm, unsigned long msiqid,
+                             unsigned long *head)
 {
-       struct pci_pbm_info *pbm = data1;
-       struct pci_sun4v_msiq_entry *base, *ep;
-       unsigned long msiqid, orig_head, head, type, err;
+       unsigned long err, limit;
 
-       msiqid = (unsigned long) data2;
-
-       head = 0xdeadbeef;
-       err = pci_sun4v_msiq_gethead(pbm->devhandle, msiqid, &head);
+       err = pci_sun4v_msiq_gethead(pbm->devhandle, msiqid, head);
        if (unlikely(err))
-               goto hv_error_get;
-
-       if (unlikely(head >= (pbm->msiq_ent_count * sizeof(struct pci_sun4v_msiq_entry))))
-               goto bad_offset;
-
-       head /= sizeof(struct pci_sun4v_msiq_entry);
-       orig_head = head;
-       base = (pbm->msi_queues + ((msiqid - pbm->msiq_first) *
-                                  (pbm->msiq_ent_count *
-                                   sizeof(struct pci_sun4v_msiq_entry))));
-       ep = &base[head];
-       while ((ep->version_type & MSIQ_TYPE_MASK) != 0) {
-               type = (ep->version_type & MSIQ_TYPE_MASK) >> MSIQ_TYPE_SHIFT;
-               if (unlikely(type != MSIQ_TYPE_MSI32 &&
-                            type != MSIQ_TYPE_MSI64))
-                       goto bad_type;
-
-               pci_sun4v_msi_setstate(pbm->devhandle,
-                                      ep->msi_data /* msi_num */,
-                                      HV_MSISTATE_IDLE);
-
-               /* Clear the entry.  */
-               ep->version_type &= ~MSIQ_TYPE_MASK;
-
-               /* Go to next entry in ring.  */
-               head++;
-               if (head >= pbm->msiq_ent_count)
-                       head = 0;
-               ep = &base[head];
-       }
+               return -ENXIO;
 
-       if (likely(head != orig_head)) {
-               /* ACK entries by updating head pointer.  */
-               head *= sizeof(struct pci_sun4v_msiq_entry);
-               err = pci_sun4v_msiq_sethead(pbm->devhandle, msiqid, head);
-               if (unlikely(err))
-                       goto hv_error_set;
-       }
-       return;
+       limit = pbm->msiq_ent_count * sizeof(struct pci_sun4v_msiq_entry);
+       if (unlikely(*head >= limit))
+               return -EFBIG;
 
-hv_error_set:
-       printk(KERN_EMERG "MSI: Hypervisor set head gives error %lu\n", err);
-       goto hv_error_cont;
+       return 0;
+}
 
-hv_error_get:
-       printk(KERN_EMERG "MSI: Hypervisor get head gives error %lu\n", err);
+static int pci_sun4v_dequeue_msi(struct pci_pbm_info *pbm,
+                                unsigned long msiqid, unsigned long *head,
+                                unsigned long *msi)
+{
+       struct pci_sun4v_msiq_entry *ep;
+       unsigned long err, type;
 
-hv_error_cont:
-       printk(KERN_EMERG "MSI: devhandle[%x] msiqid[%lx] head[%lu]\n",
-              pbm->devhandle, msiqid, head);
-       return;
+       /* Note: void pointer arithmetic, 'head' is a byte offset  */
+       ep = (pbm->msi_queues + ((msiqid - pbm->msiq_first) *
+                                (pbm->msiq_ent_count *
+                                 sizeof(struct pci_sun4v_msiq_entry))) +
+             *head);
 
-bad_offset:
-       printk(KERN_EMERG "MSI: Hypervisor gives bad offset %lx max(%lx)\n",
-              head, pbm->msiq_ent_count * sizeof(struct pci_sun4v_msiq_entry));
-       return;
+       if ((ep->version_type & MSIQ_TYPE_MASK) == 0)
+               return 0;
 
-bad_type:
-       printk(KERN_EMERG "MSI: Entry has bad type %lx\n", type);
-       return;
+       type = (ep->version_type & MSIQ_TYPE_MASK) >> MSIQ_TYPE_SHIFT;
+       if (unlikely(type != MSIQ_TYPE_MSI32 &&
+                    type != MSIQ_TYPE_MSI64))
+               return -EINVAL;
+
+       *msi = ep->msi_data;
+
+       err = pci_sun4v_msi_setstate(pbm->devhandle,
+                                    ep->msi_data /* msi_num */,
+                                    HV_MSISTATE_IDLE);
+       if (unlikely(err))
+               return -ENXIO;
+
+       /* Clear the entry.  */
+       ep->version_type &= ~MSIQ_TYPE_MASK;
+
+       (*head) += sizeof(struct pci_sun4v_msiq_entry);
+       if (*head >=
+           (pbm->msiq_ent_count * sizeof(struct pci_sun4v_msiq_entry)))
+               *head = 0;
+
+       return 1;
 }
 
-static int msi_bitmap_alloc(struct pci_pbm_info *pbm)
+static int pci_sun4v_set_head(struct pci_pbm_info *pbm, unsigned long msiqid,
+                             unsigned long head)
 {
-       unsigned long size, bits_per_ulong;
+       unsigned long err;
 
-       bits_per_ulong = sizeof(unsigned long) * 8;
-       size = (pbm->msi_num + (bits_per_ulong - 1)) & ~(bits_per_ulong - 1);
-       size /= 8;
-       BUG_ON(size % sizeof(unsigned long));
+       err = pci_sun4v_msiq_sethead(pbm->devhandle, msiqid, head);
+       if (unlikely(err))
+               return -EINVAL;
 
-       pbm->msi_bitmap = kzalloc(size, GFP_KERNEL);
-       if (!pbm->msi_bitmap)
-               return -ENOMEM;
+       return 0;
+}
 
+static int pci_sun4v_msi_setup(struct pci_pbm_info *pbm, unsigned long msiqid,
+                              unsigned long msi, int is_msi64)
+{
+       if (pci_sun4v_msi_setmsiq(pbm->devhandle, msi, msiqid,
+                                 (is_msi64 ?
+                                  HV_MSITYPE_MSI64 : HV_MSITYPE_MSI32)))
+               return -ENXIO;
+       if (pci_sun4v_msi_setstate(pbm->devhandle, msi, HV_MSISTATE_IDLE))
+               return -ENXIO;
+       if (pci_sun4v_msi_setvalid(pbm->devhandle, msi, HV_MSIVALID_VALID))
+               return -ENXIO;
        return 0;
 }
 
-static void msi_bitmap_free(struct pci_pbm_info *pbm)
+static int pci_sun4v_msi_teardown(struct pci_pbm_info *pbm, unsigned long msi)
 {
-       kfree(pbm->msi_bitmap);
-       pbm->msi_bitmap = NULL;
+       unsigned long err, msiqid;
+
+       err = pci_sun4v_msi_getmsiq(pbm->devhandle, msi, &msiqid);
+       if (err)
+               return -ENXIO;
+
+       pci_sun4v_msi_setvalid(pbm->devhandle, msi, HV_MSIVALID_INVALID);
+
+       return 0;
 }
 
-static int msi_queue_alloc(struct pci_pbm_info *pbm)
+static int pci_sun4v_msiq_alloc(struct pci_pbm_info *pbm)
 {
        unsigned long q_size, alloc_size, pages, order;
        int i;
@@ -894,234 +900,59 @@ h_error:
        return -EINVAL;
 }
 
-
-static int alloc_msi(struct pci_pbm_info *pbm)
+static void pci_sun4v_msiq_free(struct pci_pbm_info *pbm)
 {
+       unsigned long q_size, alloc_size, pages, order;
        int i;
 
-       for (i = 0; i < pbm->msi_num; i++) {
-               if (!test_and_set_bit(i, pbm->msi_bitmap))
-                       return i + pbm->msi_first;
-       }
-
-       return -ENOENT;
-}
-
-static void free_msi(struct pci_pbm_info *pbm, int msi_num)
-{
-       msi_num -= pbm->msi_first;
-       clear_bit(msi_num, pbm->msi_bitmap);
-}
-
-static int pci_sun4v_setup_msi_irq(unsigned int *virt_irq_p,
-                                  struct pci_dev *pdev,
-                                  struct msi_desc *entry)
-{
-       struct pci_pbm_info *pbm = pdev->dev.archdata.host_controller;
-       unsigned long devino, msiqid;
-       struct msi_msg msg;
-       int msi_num, err;
-
-       *virt_irq_p = 0;
-
-       msi_num = alloc_msi(pbm);
-       if (msi_num < 0)
-               return msi_num;
-
-       devino = sun4v_build_msi(pbm->devhandle, virt_irq_p,
-                                pbm->msiq_first_devino,
-                                (pbm->msiq_first_devino +
-                                 pbm->msiq_num));
-       err = -ENOMEM;
-       if (!devino)
-               goto out_err;
-
-       msiqid = ((devino - pbm->msiq_first_devino) +
-                 pbm->msiq_first);
-
-       err = -EINVAL;
-       if (pci_sun4v_msiq_setstate(pbm->devhandle, msiqid, HV_MSIQSTATE_IDLE))
-       if (err)
-               goto out_err;
-
-       if (pci_sun4v_msiq_setvalid(pbm->devhandle, msiqid, HV_MSIQ_VALID))
-               goto out_err;
-
-       if (pci_sun4v_msi_setmsiq(pbm->devhandle,
-                                 msi_num, msiqid,
-                                 (entry->msi_attrib.is_64 ?
-                                  HV_MSITYPE_MSI64 : HV_MSITYPE_MSI32)))
-               goto out_err;
-
-       if (pci_sun4v_msi_setstate(pbm->devhandle, msi_num, HV_MSISTATE_IDLE))
-               goto out_err;
-
-       if (pci_sun4v_msi_setvalid(pbm->devhandle, msi_num, HV_MSIVALID_VALID))
-               goto out_err;
-
-       pdev->dev.archdata.msi_num = msi_num;
+       for (i = 0; i < pbm->msiq_num; i++) {
+               unsigned long msiqid = pbm->msiq_first + i;
 
-       if (entry->msi_attrib.is_64) {
-               msg.address_hi = pbm->msi64_start >> 32;
-               msg.address_lo = pbm->msi64_start & 0xffffffff;
-       } else {
-               msg.address_hi = 0;
-               msg.address_lo = pbm->msi32_start;
+               (void) pci_sun4v_msiq_conf(pbm->devhandle, msiqid, 0UL, 0);
        }
-       msg.data = msi_num;
-
-       set_irq_msi(*virt_irq_p, entry);
-       write_msi_msg(*virt_irq_p, &msg);
 
-       irq_install_pre_handler(*virt_irq_p,
-                               pci_sun4v_msi_prehandler,
-                               pbm, (void *) msiqid);
+       q_size = pbm->msiq_ent_count * sizeof(struct pci_sun4v_msiq_entry);
+       alloc_size = (pbm->msiq_num * q_size);
+       order = get_order(alloc_size);
 
-       return 0;
+       pages = (unsigned long) pbm->msi_queues;
 
-out_err:
-       free_msi(pbm, msi_num);
-       sun4v_destroy_msi(*virt_irq_p);
-       *virt_irq_p = 0;
-       return err;
+       free_pages(pages, order);
 
+       pbm->msi_queues = NULL;
 }
 
-static void pci_sun4v_teardown_msi_irq(unsigned int virt_irq,
-                                      struct pci_dev *pdev)
+static int pci_sun4v_msiq_build_irq(struct pci_pbm_info *pbm,
+                                   unsigned long msiqid,
+                                   unsigned long devino)
 {
-       struct pci_pbm_info *pbm = pdev->dev.archdata.host_controller;
-       unsigned long msiqid, err;
-       unsigned int msi_num;
-
-       msi_num = pdev->dev.archdata.msi_num;
-       err = pci_sun4v_msi_getmsiq(pbm->devhandle, msi_num, &msiqid);
-       if (err) {
-               printk(KERN_ERR "%s: getmsiq gives error %lu\n",
-                      pbm->name, err);
-               return;
-       }
+       unsigned int virt_irq = sun4v_build_irq(pbm->devhandle, devino);
 
-       pci_sun4v_msi_setvalid(pbm->devhandle, msi_num, HV_MSIVALID_INVALID);
-       pci_sun4v_msiq_setvalid(pbm->devhandle, msiqid, HV_MSIQ_INVALID);
+       if (!virt_irq)
+               return -ENOMEM;
 
-       free_msi(pbm, msi_num);
+       if (pci_sun4v_msiq_setstate(pbm->devhandle, msiqid, HV_MSIQSTATE_IDLE))
+               return -EINVAL;
+       if (pci_sun4v_msiq_setvalid(pbm->devhandle, msiqid, HV_MSIQ_VALID))
+               return -EINVAL;
 
-       /* The sun4v_destroy_msi() will liberate the devino and thus the MSIQ
-        * allocation.
-        */
-       sun4v_destroy_msi(virt_irq);
+       return virt_irq;
 }
 
+static const struct sparc64_msiq_ops pci_sun4v_msiq_ops = {
+       .get_head       =       pci_sun4v_get_head,
+       .dequeue_msi    =       pci_sun4v_dequeue_msi,
+       .set_head       =       pci_sun4v_set_head,
+       .msi_setup      =       pci_sun4v_msi_setup,
+       .msi_teardown   =       pci_sun4v_msi_teardown,
+       .msiq_alloc     =       pci_sun4v_msiq_alloc,
+       .msiq_free      =       pci_sun4v_msiq_free,
+       .msiq_build_irq =       pci_sun4v_msiq_build_irq,
+};
+
 static void pci_sun4v_msi_init(struct pci_pbm_info *pbm)
 {
-       const u32 *val;
-       int len;
-
-       val = of_get_property(pbm->prom_node, "#msi-eqs", &len);
-       if (!val || len != 4)
-               goto no_msi;
-       pbm->msiq_num = *val;
-       if (pbm->msiq_num) {
-               const struct msiq_prop {
-                       u32 first_msiq;
-                       u32 num_msiq;
-                       u32 first_devino;
-               } *mqp;
-               const struct msi_range_prop {
-                       u32 first_msi;
-                       u32 num_msi;
-               } *mrng;
-               const struct addr_range_prop {
-                       u32 msi32_high;
-                       u32 msi32_low;
-                       u32 msi32_len;
-                       u32 msi64_high;
-                       u32 msi64_low;
-                       u32 msi64_len;
-               } *arng;
-
-               val = of_get_property(pbm->prom_node, "msi-eq-size", &len);
-               if (!val || len != 4)
-                       goto no_msi;
-
-               pbm->msiq_ent_count = *val;
-
-               mqp = of_get_property(pbm->prom_node,
-                                     "msi-eq-to-devino", &len);
-               if (!mqp || len != sizeof(struct msiq_prop))
-                       goto no_msi;
-
-               pbm->msiq_first = mqp->first_msiq;
-               pbm->msiq_first_devino = mqp->first_devino;
-
-               val = of_get_property(pbm->prom_node, "#msi", &len);
-               if (!val || len != 4)
-                       goto no_msi;
-               pbm->msi_num = *val;
-
-               mrng = of_get_property(pbm->prom_node, "msi-ranges", &len);
-               if (!mrng || len != sizeof(struct msi_range_prop))
-                       goto no_msi;
-               pbm->msi_first = mrng->first_msi;
-
-               val = of_get_property(pbm->prom_node, "msi-data-mask", &len);
-               if (!val || len != 4)
-                       goto no_msi;
-               pbm->msi_data_mask = *val;
-
-               val = of_get_property(pbm->prom_node, "msix-data-width", &len);
-               if (!val || len != 4)
-                       goto no_msi;
-               pbm->msix_data_width = *val;
-
-               arng = of_get_property(pbm->prom_node, "msi-address-ranges",
-                                      &len);
-               if (!arng || len != sizeof(struct addr_range_prop))
-                       goto no_msi;
-               pbm->msi32_start = ((u64)arng->msi32_high << 32) |
-                       (u64) arng->msi32_low;
-               pbm->msi64_start = ((u64)arng->msi64_high << 32) |
-                       (u64) arng->msi64_low;
-               pbm->msi32_len = arng->msi32_len;
-               pbm->msi64_len = arng->msi64_len;
-
-               if (msi_bitmap_alloc(pbm))
-                       goto no_msi;
-
-               if (msi_queue_alloc(pbm)) {
-                       msi_bitmap_free(pbm);
-                       goto no_msi;
-               }
-
-               printk(KERN_INFO "%s: MSI Queue first[%u] num[%u] count[%u] "
-                      "devino[0x%x]\n",
-                      pbm->name,
-                      pbm->msiq_first, pbm->msiq_num,
-                      pbm->msiq_ent_count,
-                      pbm->msiq_first_devino);
-               printk(KERN_INFO "%s: MSI first[%u] num[%u] mask[0x%x] "
-                      "width[%u]\n",
-                      pbm->name,
-                      pbm->msi_first, pbm->msi_num, pbm->msi_data_mask,
-                      pbm->msix_data_width);
-               printk(KERN_INFO "%s: MSI addr32[0x%lx:0x%x] "
-                      "addr64[0x%lx:0x%x]\n",
-                      pbm->name,
-                      pbm->msi32_start, pbm->msi32_len,
-                      pbm->msi64_start, pbm->msi64_len);
-               printk(KERN_INFO "%s: MSI queues at RA [%p]\n",
-                      pbm->name,
-                      pbm->msi_queues);
-       }
-       pbm->setup_msi_irq = pci_sun4v_setup_msi_irq;
-       pbm->teardown_msi_irq = pci_sun4v_teardown_msi_irq;
-
-       return;
-
-no_msi:
-       pbm->msiq_num = 0;
-       printk(KERN_INFO "%s: No MSI support.\n", pbm->name);
+       sparc64_pbm_msi_init(pbm, &pci_sun4v_msiq_ops);
 }
 #else /* CONFIG_PCI_MSI */
 static void pci_sun4v_msi_init(struct pci_pbm_info *pbm)
@@ -1186,6 +1017,8 @@ void __init sun4v_pci_init(struct device_node *dp, char *model_name)
                }
                printk("SUN4V_PCI: Registered hvapi major[%lu] minor[%lu]\n",
                       vpci_major, vpci_minor);
+
+               dma_ops = &sun4v_dma_ops;
        }
 
        prop = of_find_property(dp, "reg", NULL);
@@ -1206,7 +1039,7 @@ void __init sun4v_pci_init(struct device_node *dp, char *model_name)
                if (!page)
                        goto fatal_memory_error;
 
-               per_cpu(pci_iommu_batch, i).pglist = (u64 *) page;
+               per_cpu(iommu_batch, i).pglist = (u64 *) page;
        }
 
        p = kzalloc(sizeof(struct pci_controller_info), GFP_ATOMIC);
@@ -1225,11 +1058,6 @@ void __init sun4v_pci_init(struct device_node *dp, char *model_name)
 
        p->pbm_B.iommu = iommu;
 
-       /* Like PSYCHO and SCHIZO we have a 2GB aligned area
-        * for memory space.
-        */
-       pci_memspace_mask = 0x7fffffffUL;
-
        pci_sun4v_pbm_init(p, dp, devhandle);
        return;