pci_sun4v.c

来自「LINUX 2.6.17.4的源码」· C语言 代码 · 共 1,260 行 · 第 1/2 页

/* pci_sun4v.c: SUN4V specific PCI controller support.
 *
 * Copyright (C) 2006 David S. Miller ([e-mail])
 */

#include <linux/kernel.h>
#include <linux/types.h>
#include <linux/pci.h>
#include <linux/init.h>
#include <linux/slab.h>
#include <linux/interrupt.h>
#include <linux/percpu.h>

#include <asm/pbm.h>
#include <asm/iommu.h>
#include <asm/irq.h>
#include <asm/upa.h>
#include <asm/pstate.h>
#include <asm/oplib.h>
#include <asm/hypervisor.h>

#include "pci_impl.h"
#include "iommu_common.h"

#include "pci_sun4v.h"

#define PGLIST_NENTS	(PAGE_SIZE / sizeof(u64))

struct pci_iommu_batch {
	struct pci_dev	*pdev;		/* 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 pci_iommu_batch, pci_iommu_batch);

/* Interrupts must be disabled.  */
static inline void pci_iommu_batch_start(struct pci_dev *pdev, unsigned long prot, unsigned long entry)
{
	struct pci_iommu_batch *p = &__get_cpu_var(pci_iommu_batch);

	p->pdev		= pdev;
	p->prot		= prot;
	p->entry	= entry;
	p->npages	= 0;
}

/* Interrupts must be disabled.  */
static long pci_iommu_batch_flush(struct pci_iommu_batch *p)
{
	struct pcidev_cookie *pcp = p->pdev->sysdata;
	unsigned long devhandle = pcp->pbm->devhandle;
	unsigned long prot = p->prot;
	unsigned long entry = p->entry;
	u64 *pglist = p->pglist;
	unsigned long npages = p->npages;

	while (npages != 0) {
		long num;

		num = pci_sun4v_iommu_map(devhandle, HV_PCI_TSBID(0, entry),
					  npages, prot, __pa(pglist));
		if (unlikely(num < 0)) {
			if (printk_ratelimit())
				printk("pci_iommu_batch_flush: IOMMU map of "
				       "[%08lx:%08lx:%lx:%lx:%lx] failed with "
				       "status %ld\n",
				       devhandle, HV_PCI_TSBID(0, entry),
				       npages, prot, __pa(pglist), num);
			return -1;
		}

		entry += num;
		npages -= num;
		pglist += num;
	}

	p->entry = entry;
	p->npages = 0;

	return 0;
}

/* Interrupts must be disabled.  */
static inline long pci_iommu_batch_add(u64 phys_page)
{
	struct pci_iommu_batch *p = &__get_cpu_var(pci_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 0;
}

/* Interrupts must be disabled.  */
static inline long pci_iommu_batch_end(void)
{
	struct pci_iommu_batch *p = &__get_cpu_var(pci_iommu_batch);

	BUG_ON(p->npages >= PGLIST_NENTS);

	return pci_iommu_batch_flush(p);
}

static long pci_arena_alloc(struct pci_iommu_arena *arena, unsigned long npages)
{
	unsigned long n, i, start, end, limit;
	int pass;

	limit = arena->limit;
	start = arena->hint;
	pass = 0;

again:
	n = find_next_zero_bit(arena->map, limit, start);
	end = n + npages;
	if (unlikely(end >= limit)) {
		if (likely(pass < 1)) {
			limit = start;
			start = 0;
			pass++;
			goto again;
		} else {
			/* Scanned the whole thing, give up. */
			return -1;
		}
	}

	for (i = n; i < end; i++) {
		if (test_bit(i, arena->map)) {
			start = i + 1;
			goto again;
		}
	}

	for (i = n; i < end; i++)
		__set_bit(i, arena->map);

	arena->hint = end;

	return n;
}

static void pci_arena_free(struct pci_iommu_arena *arena, unsigned long base, unsigned long npages)
{
	unsigned long i;

	for (i = base; i < (base + npages); i++)
		__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)
{
	struct pcidev_cookie *pcp;
	struct pci_iommu *iommu;
	unsigned long flags, order, first_page, npages, n;
	void *ret;
	long entry;

	size = IO_PAGE_ALIGN(size);
	order = get_order(size);
	if (unlikely(order >= MAX_ORDER))
		return NULL;

	npages = size >> IO_PAGE_SHIFT;

	first_page = __get_free_pages(gfp, order);
	if (unlikely(first_page == 0UL))
		return NULL;

	memset((char *)first_page, 0, PAGE_SIZE << order);

	pcp = pdev->sysdata;
	iommu = pcp->pbm->iommu;

	spin_lock_irqsave(&iommu->lock, flags);
	entry = pci_arena_alloc(&iommu->arena, npages);
	spin_unlock_irqrestore(&iommu->lock, flags);

	if (unlikely(entry < 0L))
		goto arena_alloc_fail;

	*dma_addrp = (iommu->page_table_map_base +
		      (entry << IO_PAGE_SHIFT));
	ret = (void *) first_page;
	first_page = __pa(first_page);

	local_irq_save(flags);

	pci_iommu_batch_start(pdev,
			      (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));
		if (unlikely(err < 0L))
			goto iommu_map_fail;
	}

	if (unlikely(pci_iommu_batch_end() < 0L))
		goto iommu_map_fail;

	local_irq_restore(flags);

	return ret;

iommu_map_fail:
	/* Interrupts are disabled.  */
	spin_lock(&iommu->lock);
	pci_arena_free(&iommu->arena, entry, npages);
	spin_unlock_irqrestore(&iommu->lock, flags);

arena_alloc_fail:
	free_pages(first_page, order);
	return NULL;
}

static void pci_4v_free_consistent(struct pci_dev *pdev, size_t size, void *cpu, dma_addr_t dvma)
{
	struct pcidev_cookie *pcp;
	struct pci_iommu *iommu;
	unsigned long flags, order, npages, entry;
	u32 devhandle;

	npages = IO_PAGE_ALIGN(size) >> IO_PAGE_SHIFT;
	pcp = pdev->sysdata;
	iommu = pcp->pbm->iommu;
	devhandle = pcp->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);

	do {
		unsigned long num;

		num = pci_sun4v_iommu_demap(devhandle, HV_PCI_TSBID(0, entry),
					    npages);
		entry += num;
		npages -= num;
	} while (npages != 0);

	spin_unlock_irqrestore(&iommu->lock, flags);

	order = get_order(size);
	if (order < 10)
		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)
{
	struct pcidev_cookie *pcp;
	struct pci_iommu *iommu;
	unsigned long flags, npages, oaddr;
	unsigned long i, base_paddr;
	u32 bus_addr, ret;
	unsigned long prot;
	long entry;

	pcp = pdev->sysdata;
	iommu = pcp->pbm->iommu;

	if (unlikely(direction == PCI_DMA_NONE))
		goto bad;

	oaddr = (unsigned long)ptr;
	npages = IO_PAGE_ALIGN(oaddr + sz) - (oaddr & IO_PAGE_MASK);
	npages >>= IO_PAGE_SHIFT;

	spin_lock_irqsave(&iommu->lock, flags);
	entry = pci_arena_alloc(&iommu->arena, npages);
	spin_unlock_irqrestore(&iommu->lock, flags);

	if (unlikely(entry < 0L))
		goto bad;

	bus_addr = (iommu->page_table_map_base +
		    (entry << IO_PAGE_SHIFT));
	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)
		prot |= HV_PCI_MAP_ATTR_WRITE;

	local_irq_save(flags);

	pci_iommu_batch_start(pdev, prot, entry);

	for (i = 0; i < npages; i++, base_paddr += IO_PAGE_SIZE) {
		long err = pci_iommu_batch_add(base_paddr);
		if (unlikely(err < 0L))
			goto iommu_map_fail;
	}
	if (unlikely(pci_iommu_batch_end() < 0L))
		goto iommu_map_fail;

	local_irq_restore(flags);

	return ret;

bad:
	if (printk_ratelimit())
		WARN_ON(1);
	return PCI_DMA_ERROR_CODE;

iommu_map_fail:
	/* Interrupts are disabled.  */
	spin_lock(&iommu->lock);
	pci_arena_free(&iommu->arena, entry, npages);
	spin_unlock_irqrestore(&iommu->lock, flags);

	return PCI_DMA_ERROR_CODE;
}

static void pci_4v_unmap_single(struct pci_dev *pdev, dma_addr_t bus_addr, size_t sz, int direction)
{
	struct pcidev_cookie *pcp;
	struct pci_iommu *iommu;
	unsigned long flags, npages;
	long entry;
	u32 devhandle;

	if (unlikely(direction == PCI_DMA_NONE)) {
		if (printk_ratelimit())
			WARN_ON(1);
		return;
	}

	pcp = pdev->sysdata;
	iommu = pcp->pbm->iommu;
	devhandle = pcp->pbm->devhandle;

	npages = IO_PAGE_ALIGN(bus_addr + sz) - (bus_addr & IO_PAGE_MASK);
	npages >>= IO_PAGE_SHIFT;
	bus_addr &= IO_PAGE_MASK;

	spin_lock_irqsave(&iommu->lock, flags);

	entry = (bus_addr - iommu->page_table_map_base) >> IO_PAGE_SHIFT;
	pci_arena_free(&iommu->arena, entry, npages);

	do {
		unsigned long num;

		num = pci_sun4v_iommu_demap(devhandle, HV_PCI_TSBID(0, entry),
					    npages);
		entry += num;
		npages -= num;
	} while (npages != 0);

	spin_unlock_irqrestore(&iommu->lock, flags);
}

#define SG_ENT_PHYS_ADDRESS(SG)	\
	(__pa(page_address((SG)->page)) + (SG)->offset)

static inline long fill_sg(long entry, struct pci_dev *pdev,
			   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);

	for (i = 0; i < nused; i++) {
		unsigned long pteval = ~0UL;
		u32 dma_npages;

		dma_npages = ((dma_sg->dma_address & (IO_PAGE_SIZE - 1UL)) +
			      dma_sg->dma_length +
			      ((IO_PAGE_SIZE - 1UL))) >> IO_PAGE_SHIFT;
		do {
			unsigned long offset;
			signed int len;

			/* If we are here, we know we have at least one
			 * more page to map.  So walk forward until we
			 * hit a page crossing, and begin creating new
			 * mappings from that spot.
			 */
			for (;;) {
				unsigned long tmp;

				tmp = SG_ENT_PHYS_ADDRESS(sg);
				len = sg->length;
				if (((tmp ^ pteval) >> IO_PAGE_SHIFT) != 0UL) {
					pteval = tmp & IO_PAGE_MASK;
					offset = tmp & (IO_PAGE_SIZE - 1UL);
					break;
				}
				if (((tmp ^ (tmp + len - 1UL)) >> IO_PAGE_SHIFT) != 0UL) {
					pteval = (tmp + IO_PAGE_SIZE) & IO_PAGE_MASK;
					offset = 0UL;
					len -= (IO_PAGE_SIZE - (tmp & (IO_PAGE_SIZE - 1UL)));
					break;
				}
				sg++;
			}

			pteval = (pteval & IOPTE_PAGE);
			while (len > 0) {
				long err;

				err = pci_iommu_batch_add(pteval);
				if (unlikely(err < 0L))
					goto iommu_map_failed;

				pteval += IO_PAGE_SIZE;
				len -= (IO_PAGE_SIZE - offset);
				offset = 0;
				dma_npages--;
			}

			pteval = (pteval & IOPTE_PAGE) + len;
			sg++;

			/* 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 &&
			       (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++;
			}
			if ((pteval << (64 - IO_PAGE_SHIFT)) == 0UL)
				pteval = ~0UL;
		} while (dma_npages != 0);
		dma_sg++;
	}

	if (unlikely(pci_iommu_batch_end() < 0L))
		goto iommu_map_failed;

	local_irq_restore(flags);
	return 0;

iommu_map_failed:
	local_irq_restore(flags);
	return -1L;
}

static int pci_4v_map_sg(struct pci_dev *pdev, struct scatterlist *sglist, int nelems, int direction)
{
	struct pcidev_cookie *pcp;
	struct pci_iommu *iommu;
	unsigned long flags, npages, prot;
	u32 dma_base;
	struct scatterlist *sgtmp;
	long entry, err;
	int used;

	/* Fast path single entry scatterlists. */
	if (nelems == 1) {
		sglist->dma_address =
			pci_4v_map_single(pdev,
					  (page_address(sglist->page) + sglist->offset),
					  sglist->length, direction);
		if (unlikely(sglist->dma_address == PCI_DMA_ERROR_CODE))
			return 0;
		sglist->dma_length = sglist->length;
		return 1;
	}

	pcp = pdev->sysdata;
	iommu = pcp->pbm->iommu;
	
	if (unlikely(direction == PCI_DMA_NONE))
		goto bad;

	/* Step 1: Prepare scatter list. */
	npages = prepare_sg(sglist, nelems);

	/* Step 2: Allocate a cluster and context, if necessary. */
	spin_lock_irqsave(&iommu->lock, flags);
	entry = pci_arena_alloc(&iommu->arena, npages);
	spin_unlock_irqrestore(&iommu->lock, flags);

	if (unlikely(entry < 0L))
		goto bad;

	dma_base = iommu->page_table_map_base +
		(entry << IO_PAGE_SHIFT);

	/* Step 3: Normalize DMA addresses. */
	used = nelems;

	sgtmp = sglist;
	while (used && sgtmp->dma_length) {
		sgtmp->dma_address += dma_base;
		sgtmp++;
		used--;
	}
	used = nelems - used;

	/* Step 4: Create the mappings. */
	prot = HV_PCI_MAP_ATTR_READ;
	if (direction != PCI_DMA_TODEVICE)
		prot |= HV_PCI_MAP_ATTR_WRITE;

	err = fill_sg(entry, pdev, sglist, used, nelems, prot);
	if (unlikely(err < 0L))
		goto iommu_map_failed;

	return used;

bad:
	if (printk_ratelimit())
		WARN_ON(1);
	return 0;

iommu_map_failed:
	spin_lock_irqsave(&iommu->lock, flags);
	pci_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)
{
	struct pcidev_cookie *pcp;
	struct pci_iommu *iommu;
	unsigned long flags, i, npages;
	long entry;
	u32 devhandle, bus_addr;

	if (unlikely(direction == PCI_DMA_NONE)) {
		if (printk_ratelimit())
			WARN_ON(1);
	}

	pcp = pdev->sysdata;
	iommu = pcp->pbm->iommu;
	devhandle = pcp->pbm->devhandle;
	
	bus_addr = sglist->dma_address & IO_PAGE_MASK;

	for (i = 1; i < nelems; i++)
		if (sglist[i].dma_length == 0)
			break;
	i--;
	npages = (IO_PAGE_ALIGN(sglist[i].dma_address + sglist[i].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);

	do {
		unsigned long num;

		num = pci_sun4v_iommu_demap(devhandle, HV_PCI_TSBID(0, entry),
					    npages);
		entry += num;
		npages -= num;
	} while (npages != 0);

	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)
{
	/* Nothing to do... */
}

static void pci_4v_dma_sync_sg_for_cpu(struct pci_dev *pdev, struct scatterlist *sglist, int nelems, int direction)
{
	/* Nothing to do... */
}

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,
};

/* SUN4V PCI configuration space accessors. */

struct pdev_entry {
	struct pdev_entry	*next;
	u32			devhandle;
	unsigned int		bus;
	unsigned int		device;
	unsigned int		func;
};

#define PDEV_HTAB_SIZE	16
#define PDEV_HTAB_MASK	(PDEV_HTAB_SIZE - 1)
static struct pdev_entry *pdev_htab[PDEV_HTAB_SIZE];

static inline unsigned int pdev_hashfn(u32 devhandle, unsigned int bus, unsigned int device, unsigned int func)
{
	unsigned int val;

	val = (devhandle ^ (devhandle >> 4));
	val ^= bus;
	val ^= device;
	val ^= func;

	return val & PDEV_HTAB_MASK;
}

static int pdev_htab_add(u32 devhandle, unsigned int bus, unsigned int device, unsigned int func)
{
	struct pdev_entry *p = kmalloc(sizeof(*p), GFP_KERNEL);
	struct pdev_entry **slot;