Subversion Repositories shark

Rev

Rev 430 | Blame | Compare with Previous | Last modification | View Log | RSS feed

#include <linuxcomp.h>

#include <linux/pci.h>
#include <linux/slab.h>
#include <linux/module.h>

/*
 * Pool allocator ... wraps the pci_alloc_consistent page allocator, so
 * small blocks are easily used by drivers for bus mastering controllers.
 * This should probably be sharing the guts of the slab allocator.
 */


struct pci_pool {       /* the pool */
        struct list_head        page_list;
        spinlock_t              lock;
        size_t                  blocks_per_page;
        size_t                  size;
        struct pci_dev          *dev;
        size_t                  allocation;
        char                    name [32];
        wait_queue_head_t       waitq;
        struct list_head        pools;
};

struct pci_page {       /* cacheable header for 'allocation' bytes */
        struct list_head        page_list;
        void                    *vaddr;
        dma_addr_t              dma;
        unsigned                in_use;
        unsigned long           bitmap [0];
};

#define POOL_TIMEOUT_JIFFIES    ((100 /* msec */ * HZ) / 1000)
#define POOL_POISON_FREED       0xa7    /* !inuse */
#define POOL_POISON_ALLOCATED   0xa9    /* !initted */

static DECLARE_MUTEX (pools_lock);

static ssize_t
show_pools (struct device *dev, char *buf)
{
        struct pci_dev          *pdev;
        unsigned                temp, size;
        char                    *next;
        struct list_head        *i, *j;

        pdev = container_of (dev, struct pci_dev, dev);
        next = buf;
        size = PAGE_SIZE;

        temp = snprintf26(next, size, "poolinfo - 0.1\n");
        size -= temp;
        next += temp;

        down (&pools_lock);
        list_for_each (i, &pdev->pools) {
                struct pci_pool *pool;
                unsigned        pages = 0, blocks = 0;

                pool = list_entry (i, struct pci_pool, pools);

                list_for_each (j, &pool->page_list) {
                        struct pci_page *page;

                        page = list_entry (j, struct pci_page, page_list);
                        pages++;
                        blocks += page->in_use;
                }

                /* per-pool info, no real statistics yet */
                temp = snprintf26(next, size, "%-16s %4u %4Zu %4Zu %2u\n",
                                pool->name,
                                blocks, (unsigned int)(pages * pool->blocks_per_page),
                                (unsigned int)(pool->size), pages);
                size -= temp;
                next += temp;
        }
        up (&pools_lock);

        return PAGE_SIZE - size;
}
static DEVICE_ATTR (pools, S_IRUGO, show_pools, NULL);

/**
 * pci_pool_create - Creates a pool of pci consistent memory blocks, for dma.
 * @name: name of pool, for diagnostics
 * @pdev: pci device that will be doing the DMA
 * @size: size of the blocks in this pool.
 * @align: alignment requirement for blocks; must be a power of two
 * @allocation: returned blocks won't cross this boundary (or zero)
 * Context: !in_interrupt()
 *
 * Returns a pci allocation pool with the requested characteristics, or
 * null if one can't be created.  Given one of these pools, pci_pool_alloc()
 * may be used to allocate memory.  Such memory will all have "consistent"
 * DMA mappings, accessible by the device and its driver without using
 * cache flushing primitives.  The actual size of blocks allocated may be
 * larger than requested because of alignment.
 *
 * If allocation is nonzero, objects returned from pci_pool_alloc() won't
 * cross that size boundary.  This is useful for devices which have
 * addressing restrictions on individual DMA transfers, such as not crossing
 * boundaries of 4KBytes.
 */

struct pci_pool *
pci_pool_create (const char *name, struct pci_dev *pdev,
        size_t size, size_t align, size_t allocation)
{
        struct pci_pool         *retval;

        if (align == 0)
                align = 1;
        if (size == 0)
                return 0;
        else if (size < align)
                size = align;
        else if ((size % align) != 0) {
                size += align + 1;
                size &= ~(align - 1);
        }

        if (allocation == 0) {
                if (PAGE_SIZE < size)
                        allocation = size;
                else
                        allocation = PAGE_SIZE;
                // FIXME: round up for less fragmentation
        } else if (allocation < size)
                return 0;

        if (!(retval = kmalloc (sizeof *retval, SLAB_KERNEL)))
                return retval;

        strlcpy (retval->name, name, sizeof retval->name);

        retval->dev = pdev;

        INIT_LIST_HEAD (&retval->page_list);
        spin_lock_init (&retval->lock);
        retval->size = size;
        retval->allocation = allocation;
        retval->blocks_per_page = allocation / size;
        init_waitqueue_head (&retval->waitq);

        if (pdev) {
                down (&pools_lock);
                if (list_empty (&pdev->pools))
                        device_create_file (&pdev->dev, &dev_attr_pools);
                /* note:  not currently insisting "name" be unique */
                list_add (&retval->pools, &pdev->pools);
                up (&pools_lock);
        } else
                INIT_LIST_HEAD (&retval->pools);

        return retval;
}


static struct pci_page *
pool_alloc_page (struct pci_pool *pool, int mem_flags)
{
        struct pci_page *page;
        int             mapsize;

        mapsize = pool->blocks_per_page;
        mapsize = (mapsize + BITS_PER_LONG - 1) / BITS_PER_LONG;
        mapsize *= sizeof (long);

        page = (struct pci_page *) kmalloc (mapsize + sizeof *page, mem_flags);
        if (!page)
                return 0;
        page->vaddr = pci_alloc_consistent (pool->dev,
                                            pool->allocation,
                                            &page->dma);
        if (page->vaddr) {
                memset (page->bitmap, 0xff, mapsize);   // bit set == free
#ifdef  CONFIG_DEBUG_SLAB
                memset (page->vaddr, POOL_POISON_FREED, pool->allocation);
#endif
                list_add (&page->page_list, &pool->page_list);
                page->in_use = 0;
        } else {
                kfree (page);
                page = 0;
        }
        return page;
}


static inline int
is_page_busy (int blocks, unsigned long *bitmap)
{
        while (blocks > 0) {
                if (*bitmap++ != ~0UL)
                        return 1;
                blocks -= BITS_PER_LONG;
        }
        return 0;
}

static void
pool_free_page (struct pci_pool *pool, struct pci_page *page)
{
        dma_addr_t      dma = page->dma;

#ifdef  CONFIG_DEBUG_SLAB
        memset (page->vaddr, POOL_POISON_FREED, pool->allocation);
#endif
        pci_free_consistent (pool->dev, pool->allocation, page->vaddr, dma);
        list_del (&page->page_list);
        kfree (page);
}


/**
 * pci_pool_destroy - destroys a pool of pci memory blocks.
 * @pool: pci pool that will be destroyed
 * Context: !in_interrupt()
 *
 * Caller guarantees that no more memory from the pool is in use,
 * and that nothing will try to use the pool after this call.
 */

void
pci_pool_destroy (struct pci_pool *pool)
{
        down (&pools_lock);
        list_del (&pool->pools);
        if (pool->dev && list_empty (&pool->dev->pools))
                device_remove_file (&pool->dev->dev, &dev_attr_pools);
        up (&pools_lock);

        while (!list_empty (&pool->page_list)) {
                struct pci_page         *page;
                page = list_entry (pool->page_list.next,
                                struct pci_page, page_list);
                if (is_page_busy (pool->blocks_per_page, page->bitmap)) {
                        printk (KERN_ERR "pci_pool_destroy %s/%s, %p busy\n",
                                pool->dev ? pci_name(pool->dev) : NULL,
                                pool->name, page->vaddr);
                        /* leak the still-in-use consistent memory */
                        list_del (&page->page_list);
                        kfree (page);
                } else
                        pool_free_page (pool, page);
        }

        kfree (pool);
}


/**
 * pci_pool_alloc - get a block of consistent memory
 * @pool: pci pool that will produce the block
 * @mem_flags: SLAB_KERNEL or SLAB_ATOMIC
 * @handle: pointer to dma address of block
 *
 * This returns the kernel virtual address of a currently unused block,
 * and reports its dma address through the handle.
 * If such a memory block can't be allocated, null is returned.
 */

void *
pci_pool_alloc (struct pci_pool *pool, int mem_flags, dma_addr_t *handle)
{
        unsigned long           flags;
        struct list_head        *entry;
        struct pci_page         *page;
        int                     map, block;
        size_t                  offset;
        void                    *retval;

restart:
        spin_lock_irqsave (&pool->lock, flags);
        list_for_each (entry, &pool->page_list) {
                int             i;
                page = list_entry (entry, struct pci_page, page_list);
                /* only cachable accesses here ... */
                for (map = 0, i = 0;
                                i < pool->blocks_per_page;
                                i += BITS_PER_LONG, map++) {
                        if (page->bitmap [map] == 0)
                                continue;
                        block = ffz (~ page->bitmap [map]);
                        if ((i + block) < pool->blocks_per_page) {
                                clear_bit (block, &page->bitmap [map]);
                                offset = (BITS_PER_LONG * map) + block;
                                offset *= pool->size;
                                goto ready;
                        }
                }
        }
        if (!(page = pool_alloc_page (pool, SLAB_ATOMIC))) {
                if (mem_flags == SLAB_KERNEL) {
                        DECLARE_WAITQUEUE (wait, current);

                        current->state = TASK_INTERRUPTIBLE;
                        add_wait_queue (&pool->waitq, &wait);
                        spin_unlock_irqrestore (&pool->lock, flags);

                        schedule_timeout (POOL_TIMEOUT_JIFFIES);

                        remove_wait_queue (&pool->waitq, &wait);
                        goto restart;
                }
                retval = 0;
                goto done;
        }

        clear_bit (0, &page->bitmap [0]);
        offset = 0;
ready:
        page->in_use++;
        retval = offset + page->vaddr;
        *handle = offset + page->dma;
#ifdef  CONFIG_DEBUG_SLAB
        memset (retval, POOL_POISON_ALLOCATED, pool->size);
#endif
done:
        spin_unlock_irqrestore (&pool->lock, flags);
        return retval;
}


static struct pci_page *
pool_find_page (struct pci_pool *pool, dma_addr_t dma)
{
        unsigned long           flags;
        struct list_head        *entry;
        struct pci_page         *page;

        spin_lock_irqsave (&pool->lock, flags);
        list_for_each (entry, &pool->page_list) {
                page = list_entry (entry, struct pci_page, page_list);
                if (dma < page->dma)
                        continue;
                if (dma < (page->dma + pool->allocation))
                        goto done;
        }
        page = 0;
done:
        spin_unlock_irqrestore (&pool->lock, flags);
        return page;
}


/**
 * pci_pool_free - put block back into pci pool
 * @pool: the pci pool holding the block
 * @vaddr: virtual address of block
 * @dma: dma address of block
 *
 * Caller promises neither device nor driver will again touch this block
 * unless it is first re-allocated.
 */

void
pci_pool_free (struct pci_pool *pool, void *vaddr, dma_addr_t dma)
{
        struct pci_page         *page;
        unsigned long           flags;
        int                     map, block;

        if ((page = pool_find_page (pool, dma)) == 0) {
                printk (KERN_ERR "pci_pool_free %s/%s, %p/%lx (bad dma)\n",
                        pool->dev ? pci_name(pool->dev) : NULL,
                        pool->name, vaddr, (unsigned long) dma);
                return;
        }

        block = dma - page->dma;
        block /= pool->size;
        map = block / BITS_PER_LONG;
        block %= BITS_PER_LONG;

#ifdef  CONFIG_DEBUG_SLAB
        if (((dma - page->dma) + (void *)page->vaddr) != vaddr) {
                printk (KERN_ERR "pci_pool_free %s/%s, %p (bad vaddr)/%Lx\n",
                        pool->dev ? pci_name(pool->dev) : NULL,
                        pool->name, vaddr, (unsigned long long) dma);
                return;
        }
        if (page->bitmap [map] & (1UL << block)) {
                printk (KERN_ERR "pci_pool_free %s/%s, dma %Lx already free\n",
                        pool->dev ? pci_name(pool->dev) : NULL,
                        pool->name, (unsigned long long)dma);
                return;
        }
        memset (vaddr, POOL_POISON_FREED, pool->size);
#endif

        spin_lock_irqsave (&pool->lock, flags);
        page->in_use--;
        set_bit (block, &page->bitmap [map]);
        if (waitqueue_active (&pool->waitq))
                wake_up (&pool->waitq);
        /*
         * Resist a temptation to do
         *    if (!is_page_busy(bpp, page->bitmap)) pool_free_page(pool, page);
         * it is not interrupt safe. Better have empty pages hang around.
         */

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


EXPORT_SYMBOL (pci_pool_create);
EXPORT_SYMBOL (pci_pool_destroy);
EXPORT_SYMBOL (pci_pool_alloc);
EXPORT_SYMBOL (pci_pool_free);