ccio-dma.c

这个linux源代码是很全面的~基本完整了~使用c编译的~由于时间问题我没有亲自测试~但就算用来做参考资料也是非常好的

#define HINT_UDPATE_ENB 0x08UL  /* not used/supported by U2 */
#define HINT_PREFETCH   0x10UL	/* for outbound pages which are not SAFE */


/*
** Use direction (ie PCI_DMA_TODEVICE) to pick hint.
** ccio_alloc_consistent() depends on this to get SAFE_DMA
** when it passes in BIDIRECTIONAL flag.
*/
static u32 hint_lookup[] = {
	[PCI_DMA_BIDIRECTIONAL]  HINT_STOP_MOST | HINT_SAFE_DMA | IOPDIR_VALID,
	[PCI_DMA_TODEVICE]       HINT_STOP_MOST | HINT_PREFETCH | IOPDIR_VALID,
	[PCI_DMA_FROMDEVICE]     HINT_STOP_MOST | IOPDIR_VALID,
	[PCI_DMA_NONE]           0,            /* not valid */
};

/**
 * ccio_io_pdir_entry - Initialize an I/O Pdir.
 * @pdir_ptr: A pointer into I/O Pdir.
 * @sid: The Space Identifier.
 * @vba: The virtual address.
 * @hints: The DMA Hint.
 *
 * Given a virtual address (vba, arg2) and space id, (sid, arg1),
 * load the I/O PDIR entry pointed to by pdir_ptr (arg0). Each IO Pdir
 * entry consists of 8 bytes as shown below (MSB == bit 0):
 *
 *
 * WORD 0:
 * +------+----------------+-----------------------------------------------+
 * | Phys | Virtual Index  |               Phys                            |
 * | 0:3  |     0:11       |               4:19                            |
 * |4 bits|   12 bits      |              16 bits                          |
 * +------+----------------+-----------------------------------------------+
 * WORD 1:
 * +-----------------------+-----------------------------------------------+
 * |      Phys    |  Rsvd  | Prefetch |Update |Rsvd  |Lock  |Safe  |Valid  |
 * |     20:39    |        | Enable   |Enable |      |Enable|DMA   |       |
 * |    20 bits   | 5 bits | 1 bit    |1 bit  |2 bits|1 bit |1 bit |1 bit  |
 * +-----------------------+-----------------------------------------------+
 *
 * The virtual index field is filled with the results of the LCI
 * (Load Coherence Index) instruction.  The 8 bits used for the virtual
 * index are bits 12:19 of the value returned by LCI.
 */ 
void CCIO_INLINE
ccio_io_pdir_entry(u64 *pdir_ptr, space_t sid, void * vba, unsigned long hints)
{
	register unsigned long pa = (volatile unsigned long) vba;
	register unsigned long ci; /* coherent index */

	/* We currently only support kernel addresses */
	ASSERT(sid == KERNEL_SPACE);

	mtsp(sid,1);

	/*
	** WORD 1 - low order word
	** "hints" parm includes the VALID bit!
	** "dep" clobbers the physical address offset bits as well.
	*/
	pa = virt_to_phys(vba);
	asm volatile("depw  %1,31,12,%0" : "+r" (pa) : "r" (hints));
	((u32 *)pdir_ptr)[1] = (u32) pa;

	/*
	** WORD 0 - high order word
	*/

#ifdef __LP64__
	/*
	** get bits 12:15 of physical address
	** shift bits 16:31 of physical address
	** and deposit them
	*/
	asm volatile ("extrd,u %1,15,4,%0" : "=r" (ci) : "r" (pa));
	asm volatile ("extrd,u %1,31,16,%0" : "+r" (pa) : "r" (pa));
	asm volatile ("depd  %1,35,4,%0" : "+r" (pa) : "r" (ci));
#else
	pa = 0;
#endif
	/*
	** get CPU coherency index bits
	** Grab virtual index [0:11]
	** Deposit virt_idx bits into I/O PDIR word
	*/
	asm volatile ("lci 0(%%sr1, %1), %0" : "=r" (ci) : "r" (vba));
	asm volatile ("extru %1,19,12,%0" : "+r" (ci) : "r" (ci));
	asm volatile ("depw  %1,15,12,%0" : "+r" (pa) : "r" (ci));

	((u32 *)pdir_ptr)[0] = (u32) pa;


	/* FIXME: PCX_W platforms don't need FDC/SYNC. (eg C360)
	**        PCX-U/U+ do. (eg C200/C240)
	**        PCX-T'? Don't know. (eg C110 or similar K-class)
	**
	** See PDC_MODEL/option 0/SW_CAP word for "Non-coherent IO-PDIR bit".
	** Hopefully we can patch (NOP) these out at boot time somehow.
	**
	** "Since PCX-U employs an offset hash that is incompatible with
	** the real mode coherence index generation of U2, the PDIR entry
	** must be flushed to memory to retain coherence."
	*/
	asm volatile("fdc 0(%0)" : : "r" (pdir_ptr));
	asm volatile("sync");
}

/**
 * ccio_clear_io_tlb - Remove stale entries from the I/O TLB.
 * @ioc: The I/O Controller.
 * @iovp: The I/O Virtual Page.
 * @byte_cnt: The requested number of bytes to be freed from the I/O Pdir.
 *
 * Purge invalid I/O PDIR entries from the I/O TLB.
 *
 * FIXME: Can we change the byte_cnt to pages_mapped?
 */
static CCIO_INLINE void
ccio_clear_io_tlb(struct ioc *ioc, dma_addr_t iovp, size_t byte_cnt)
{
	u32 chain_size = 1 << ioc->chainid_shift;

	iovp &= IOVP_MASK;	/* clear offset bits, just want pagenum */
	byte_cnt += chain_size;

	while(byte_cnt > chain_size) {
		WRITE_U32(CMD_TLB_PURGE | iovp, &ioc->ioc_hpa->io_command);
		iovp += chain_size;
		byte_cnt -= chain_size;
      }
}

/**
 * ccio_mark_invalid - Mark the I/O Pdir entries invalid.
 * @ioc: The I/O Controller.
 * @iova: The I/O Virtual Address.
 * @byte_cnt: The requested number of bytes to be freed from the I/O Pdir.
 *
 * Mark the I/O Pdir entries invalid and blow away the corresponding I/O
 * TLB entries.
 *
 * FIXME: at some threshhold it might be "cheaper" to just blow
 *        away the entire I/O TLB instead of individual entries.
 *
 * FIXME: Uturn has 256 TLB entries. We don't need to purge every
 *        PDIR entry - just once for each possible TLB entry.
 *        (We do need to maker I/O PDIR entries invalid regardless).
 *
 * FIXME: Can we change byte_cnt to pages_mapped?
 */ 
static CCIO_INLINE void
ccio_mark_invalid(struct ioc *ioc, dma_addr_t iova, size_t byte_cnt)
{
	u32 iovp = (u32)CCIO_IOVP(iova);
	size_t saved_byte_cnt;

	/* round up to nearest page size */
	saved_byte_cnt = byte_cnt = ROUNDUP(byte_cnt, IOVP_SIZE);

	while(byte_cnt > 0) {
		/* invalidate one page at a time */
		unsigned int idx = PDIR_INDEX(iovp);
		char *pdir_ptr = (char *) &(ioc->pdir_base[idx]);

		ASSERT(idx < (ioc->pdir_size / sizeof(u64)));
		pdir_ptr[7] = 0;	/* clear only VALID bit */ 
		/*
		** FIXME: PCX_W platforms don't need FDC/SYNC. (eg C360)
		**   PCX-U/U+ do. (eg C200/C240)
		** See PDC_MODEL/option 0/SW_CAP for "Non-coherent IO-PDIR bit".
		**
		** Hopefully someone figures out how to patch (NOP) the
		** FDC/SYNC out at boot time.
		*/
		asm volatile("fdc 0(%0)" : : "r" (pdir_ptr[7]));

		iovp     += IOVP_SIZE;
		byte_cnt -= IOVP_SIZE;
	}

	asm volatile("sync");
	ccio_clear_io_tlb(ioc, CCIO_IOVP(iova), saved_byte_cnt);
}

/****************************************************************
**
**          CCIO dma_ops
**
*****************************************************************/

/**
 * ccio_dma_supported - Verify the IOMMU supports the DMA address range.
 * @dev: The PCI device.
 * @mask: A bit mask describing the DMA address range of the device.
 *
 * This function implements the pci_dma_supported function.
 */
static int 
ccio_dma_supported(struct pci_dev *dev, u64 mask)
{
	if(dev == NULL) {
		printk(KERN_ERR MODULE_NAME ": EISA/ISA/et al not supported\n");
		BUG();
		return 0;
	}

	dev->dma_mask = mask;   /* save it */

	/* only support 32-bit devices (ie PCI/GSC) */
	return (int)(mask == 0xffffffffUL);
}

/**
 * ccio_map_single - Map an address range into the IOMMU.
 * @dev: The PCI device.
 * @addr: The start address of the DMA region.
 * @size: The length of the DMA region.
 * @direction: The direction of the DMA transaction (to/from device).
 *
 * This function implements the pci_map_single function.
 */
static dma_addr_t 
ccio_map_single(struct pci_dev *dev, void *addr, size_t size, int direction)
{
	int idx;
	struct ioc *ioc;
	unsigned long flags;
	dma_addr_t iovp;
	dma_addr_t offset;
	u64 *pdir_start;
	unsigned long hint = hint_lookup[direction];

	ASSERT(dev);
	ASSERT(dev->sysdata);
	ASSERT(HBA_DATA(dev->sysdata)->iommu);
	ioc = GET_IOC(dev);

	ASSERT(size > 0);

	/* save offset bits */
	offset = ((unsigned long) addr) & ~IOVP_MASK;

	/* round up to nearest IOVP_SIZE */
	size = ROUNDUP(size + offset, IOVP_SIZE);
	spin_lock_irqsave(&ioc->res_lock, flags);

#ifdef CONFIG_PROC_FS
	ioc->msingle_calls++;
	ioc->msingle_pages += size >> IOVP_SHIFT;
#endif

	idx = ccio_alloc_range(ioc, (size >> IOVP_SHIFT));
	iovp = (dma_addr_t)MKIOVP(idx);

	pdir_start = &(ioc->pdir_base[idx]);

	DBG_RUN("%s() 0x%p -> 0x%lx size: %0x%x\n",
		__FUNCTION__, addr, (long)iovp | offset, size);

	/* If not cacheline aligned, force SAFE_DMA on the whole mess */
	if((size % L1_CACHE_BYTES) || ((unsigned long)addr % L1_CACHE_BYTES))
		hint |= HINT_SAFE_DMA;

	while(size > 0) {
		ccio_io_pdir_entry(pdir_start, KERNEL_SPACE, addr, hint);

		DBG_RUN(" pdir %p %08x%08x\n",
			pdir_start,
			(u32) (((u32 *) pdir_start)[0]),
			(u32) (((u32 *) pdir_start)[1]));
		++pdir_start;
		addr += IOVP_SIZE;
		size -= IOVP_SIZE;
	}

	spin_unlock_irqrestore(&ioc->res_lock, flags);

	/* form complete address */
	return CCIO_IOVA(iovp, offset);
}

/**
 * ccio_unmap_single - Unmap an address range from the IOMMU.
 * @dev: The PCI device.
 * @addr: The start address of the DMA region.
 * @size: The length of the DMA region.
 * @direction: The direction of the DMA transaction (to/from device).
 *
 * This function implements the pci_unmap_single function.
 */
static void 
ccio_unmap_single(struct pci_dev *dev, dma_addr_t iova, size_t size, 
		  int direction)
{
	struct ioc *ioc;
	unsigned long flags; 
	dma_addr_t offset = iova & ~IOVP_MASK;
	
	ASSERT(dev);
	ASSERT(dev->sysdata);
	ASSERT(HBA_DATA(dev->sysdata)->iommu);
	ioc = GET_IOC(dev);

	DBG_RUN("%s() iovp 0x%lx/%x\n",
		__FUNCTION__, (long)iova, size);

	iova ^= offset;        /* clear offset bits */
	size += offset;
	size = ROUNDUP(size, IOVP_SIZE);

	spin_lock_irqsave(&ioc->res_lock, flags);

#ifdef CONFIG_PROC_FS
	ioc->usingle_calls++;
	ioc->usingle_pages += size >> IOVP_SHIFT;
#endif

	ccio_mark_invalid(ioc, iova, size);
	ccio_free_range(ioc, iova, (size >> IOVP_SHIFT));
	spin_unlock_irqrestore(&ioc->res_lock, flags);
}

/**
 * ccio_alloc_consistent - Allocate a consistent DMA mapping.
 * @dev: The PCI device.
 * @size: The length of the DMA region.
 * @dma_handle: The DMA address handed back to the device (not the cpu).
 *
 * This function implements the pci_alloc_consistent function.
 */
static void * 
ccio_alloc_consistent(struct pci_dev *dev, size_t size, dma_addr_t *dma_handle)
{
      void *ret;
#if 0
/* GRANT Need to establish hierarchy for non-PCI devs as well
** and then provide matching gsc_map_xxx() functions for them as well.
*/
	if(!hwdev) {
		/* only support PCI */
		*dma_handle = 0;
		return 0;
	}
#endif
        ret = (void *) __get_free_pages(GFP_ATOMIC, get_order(size));

	if (ret) {
		memset(ret, 0, size);
		*dma_handle = ccio_map_single(dev, ret, size, PCI_DMA_BIDIRECTIONAL);
	}

	return ret;
}

/**
 * ccio_free_consistent - Free a consistent DMA mapping.
 * @dev: The PCI device.
 * @size: The length of the DMA region.
 * @cpu_addr: The cpu address returned from the ccio_alloc_consistent.
 * @dma_handle: The device address returned from the ccio_alloc_consistent.
 *
 * This function implements the pci_free_consistent function.
 */
static void 
ccio_free_consistent(struct pci_dev *dev, size_t size, void *cpu_addr, 
		     dma_addr_t dma_handle)
{
	ccio_unmap_single(dev, dma_handle, size, 0);
	free_pages((unsigned long)cpu_addr, get_order(size));
}

/*
** Since 0 is a valid pdir_base index value, can't use that
** to determine if a value is valid or not. Use a flag to indicate
** the SG list entry contains a valid pdir index.
*/
#define PIDE_FLAG 0x80000000UL

/**
 * ccio_fill_pdir - Insert coalesced scatter/gather chunks into the I/O Pdir.
 * @ioc: The I/O Controller.
 * @startsg: The scatter/gather list of coalesced chunks.
 * @nents: The number of entries in the scatter/gather list.
 * @hint: The DMA Hint.
 *
 * This function inserts the coalesced scatter/gather list chunks into the
 * I/O Controller's I/O Pdir.
 */ 
static CCIO_INLINE int
ccio_fill_pdir(struct ioc *ioc, struct scatterlist *startsg, int nents, 
	       unsigned long hint)
{
	struct scatterlist *dma_sg = startsg;	/* pointer to current DMA */
	int n_mappings = 0;
	u64 *pdirp = 0;
	unsigned long dma_offset = 0;

	dma_sg--;
	while (nents-- > 0) {
		int cnt = sg_dma_len(startsg);
		sg_dma_len(startsg) = 0;

		DBG_RUN_SG(" %d : %08lx/%05x %p/%05x\n", nents,