sba_iommu.c

h内核

	/*
  	** Set all the PDIR entries valid w/ the spill page as the target
	*/
	for (index = 0 ; index < (ioc->pdir_size / PDIR_ENTRY_SIZE) ; index++)
		((u64 *)ioc->pdir_base)[index] = (0x80000000000000FF | prefetch_spill_page);
#endif

	/* Clear I/O TLB of any possible entries */
	WRITE_REG(ioc->ibase | (get_iovp_order(ioc->iov_size) + iovp_shift), ioc->ioc_hpa + IOC_PCOM);
	READ_REG(ioc->ioc_hpa + IOC_PCOM);

	/* Enable IOVA translation */
	WRITE_REG(ioc->ibase | 1, ioc->ioc_hpa + IOC_IBASE);
	READ_REG(ioc->ioc_hpa + IOC_IBASE);
}

static void __init
ioc_resource_init(struct ioc *ioc)
{
	spin_lock_init(&ioc->res_lock);
#if DELAYED_RESOURCE_CNT > 0
	spin_lock_init(&ioc->saved_lock);
#endif

	/* resource map size dictated by pdir_size */
	ioc->res_size = ioc->pdir_size / PDIR_ENTRY_SIZE; /* entries */
	ioc->res_size >>= 3;  /* convert bit count to byte count */
	DBG_INIT("%s() res_size 0x%x\n", __FUNCTION__, ioc->res_size);

	ioc->res_map = (char *) __get_free_pages(GFP_KERNEL,
						 get_order(ioc->res_size));
	if (!ioc->res_map)
		panic(PFX "Couldn't allocate resource map\n");

	memset(ioc->res_map, 0, ioc->res_size);
	/* next available IOVP - circular search */
	ioc->res_hint = (unsigned long *) ioc->res_map;

#ifdef ASSERT_PDIR_SANITY
	/* Mark first bit busy - ie no IOVA 0 */
	ioc->res_map[0] = 0x1;
	ioc->pdir_base[0] = 0x8000000000000000ULL | ZX1_SBA_IOMMU_COOKIE;
#endif
#ifdef FULL_VALID_PDIR
	/* Mark the last resource used so we don't prefetch beyond IOVA space */
	ioc->res_map[ioc->res_size - 1] |= 0x80UL; /* res_map is chars */
	ioc->pdir_base[(ioc->pdir_size / PDIR_ENTRY_SIZE) - 1] = (0x80000000000000FF
							      | prefetch_spill_page);
#endif

	DBG_INIT("%s() res_map %x %p\n", __FUNCTION__,
		 ioc->res_size, (void *) ioc->res_map);
}

static void __init
ioc_sac_init(struct ioc *ioc)
{
	struct pci_dev *sac = NULL;
	struct pci_controller *controller = NULL;

	/*
	 * pci_alloc_coherent() must return a DMA address which is
	 * SAC (single address cycle) addressable, so allocate a
	 * pseudo-device to enforce that.
	 */
	sac = kmalloc(sizeof(*sac), GFP_KERNEL);
	if (!sac)
		panic(PFX "Couldn't allocate struct pci_dev");
	memset(sac, 0, sizeof(*sac));

	controller = kmalloc(sizeof(*controller), GFP_KERNEL);
	if (!controller)
		panic(PFX "Couldn't allocate struct pci_controller");
	memset(controller, 0, sizeof(*controller));

	controller->iommu = ioc;
	sac->sysdata = controller;
	sac->dma_mask = 0xFFFFFFFFUL;
#ifdef CONFIG_PCI
	sac->dev.bus = &pci_bus_type;
#endif
	ioc->sac_only_dev = sac;
}

static void __init
ioc_zx1_init(struct ioc *ioc)
{
	unsigned long rope_config;
	unsigned int i;

	if (ioc->rev < 0x20)
		panic(PFX "IOC 2.0 or later required for IOMMU support\n");

	/* 38 bit memory controller + extra bit for range displaced by MMIO */
	ioc->dma_mask = (0x1UL << 39) - 1;

	/*
	** Clear ROPE(N)_CONFIG AO bit.
	** Disables "NT Ordering" (~= !"Relaxed Ordering")
	** Overrides bit 1 in DMA Hint Sets.
	** Improves netperf UDP_STREAM by ~10% for tg3 on bcm5701.
	*/
	for (i=0; i<(8*8); i+=8) {
		rope_config = READ_REG(ioc->ioc_hpa + IOC_ROPE0_CFG + i);
		rope_config &= ~IOC_ROPE_AO;
		WRITE_REG(rope_config, ioc->ioc_hpa + IOC_ROPE0_CFG + i);
	}
}

typedef void (initfunc)(struct ioc *);

struct ioc_iommu {
	u32 func_id;
	char *name;
	initfunc *init;
};

static struct ioc_iommu ioc_iommu_info[] __initdata = {
	{ ZX1_IOC_ID, "zx1", ioc_zx1_init },
	{ SX1000_IOC_ID, "sx1000", NULL },
};

static struct ioc * __init
ioc_init(u64 hpa, void *handle)
{
	struct ioc *ioc;
	struct ioc_iommu *info;

	ioc = kmalloc(sizeof(*ioc), GFP_KERNEL);
	if (!ioc)
		return NULL;

	memset(ioc, 0, sizeof(*ioc));

	ioc->next = ioc_list;
	ioc_list = ioc;

	ioc->handle = handle;
	ioc->ioc_hpa = ioremap(hpa, 0x1000);

	ioc->func_id = READ_REG(ioc->ioc_hpa + IOC_FUNC_ID);
	ioc->rev = READ_REG(ioc->ioc_hpa + IOC_FCLASS) & 0xFFUL;
	ioc->dma_mask = 0xFFFFFFFFFFFFFFFFUL;	/* conservative */

	for (info = ioc_iommu_info; info < ioc_iommu_info + ARRAY_SIZE(ioc_iommu_info); info++) {
		if (ioc->func_id == info->func_id) {
			ioc->name = info->name;
			if (info->init)
				(info->init)(ioc);
		}
	}

	iovp_size = (1 << iovp_shift);
	iovp_mask = ~(iovp_size - 1);

	DBG_INIT("%s: PAGE_SIZE %ldK, iovp_size %ldK\n", __FUNCTION__,
		PAGE_SIZE >> 10, iovp_size >> 10);

	if (!ioc->name) {
		ioc->name = kmalloc(24, GFP_KERNEL);
		if (ioc->name)
			sprintf((char *) ioc->name, "Unknown (%04x:%04x)",
				ioc->func_id & 0xFFFF, (ioc->func_id >> 16) & 0xFFFF);
		else
			ioc->name = "Unknown";
	}

	ioc_iova_init(ioc);
	ioc_resource_init(ioc);
	ioc_sac_init(ioc);

	if ((long) ~iovp_mask > (long) ia64_max_iommu_merge_mask)
		ia64_max_iommu_merge_mask = ~iovp_mask;

	printk(KERN_INFO PFX
		"%s %d.%d HPA 0x%lx IOVA space %dMb at 0x%lx\n",
		ioc->name, (ioc->rev >> 4) & 0xF, ioc->rev & 0xF,
		hpa, ioc->iov_size >> 20, ioc->ibase);

	return ioc;
}



/**************************************************************************
**
**   SBA initialization code (HW and SW)
**
**   o identify SBA chip itself
**   o FIXME: initialize DMA hints for reasonable defaults
**
**************************************************************************/

#ifdef CONFIG_PROC_FS
static void *
ioc_start(struct seq_file *s, loff_t *pos)
{
	struct ioc *ioc;
	loff_t n = *pos;

	for (ioc = ioc_list; ioc; ioc = ioc->next)
		if (!n--)
			return ioc;

	return NULL;
}

static void *
ioc_next(struct seq_file *s, void *v, loff_t *pos)
{
	struct ioc *ioc = v;

	++*pos;
	return ioc->next;
}

static void
ioc_stop(struct seq_file *s, void *v)
{
}

static int
ioc_show(struct seq_file *s, void *v)
{
	struct ioc *ioc = v;
	unsigned long *res_ptr = (unsigned long *)ioc->res_map;
	int i, used = 0;

	seq_printf(s, "Hewlett Packard %s IOC rev %d.%d\n",
		ioc->name, ((ioc->rev >> 4) & 0xF), (ioc->rev & 0xF));
#ifdef CONFIG_NUMA
	if (ioc->node != MAX_NUMNODES)
		seq_printf(s, "NUMA node       : %d\n", ioc->node);
#endif
	seq_printf(s, "IOVA size       : %ld MB\n", ((ioc->pdir_size >> 3) * iovp_size)/(1024*1024));
	seq_printf(s, "IOVA page size  : %ld kb\n", iovp_size/1024);

	for (i = 0; i < (ioc->res_size / sizeof(unsigned long)); ++i, ++res_ptr)
		used += hweight64(*res_ptr);

	seq_printf(s, "PDIR size       : %d entries\n", ioc->pdir_size >> 3);
	seq_printf(s, "PDIR used       : %d entries\n", used);

#ifdef PDIR_SEARCH_TIMING
	{
		unsigned long i = 0, avg = 0, min, max;
		min = max = ioc->avg_search[0];
		for (i = 0; i < SBA_SEARCH_SAMPLE; i++) {
			avg += ioc->avg_search[i];
			if (ioc->avg_search[i] > max) max = ioc->avg_search[i];
			if (ioc->avg_search[i] < min) min = ioc->avg_search[i];
		}
		avg /= SBA_SEARCH_SAMPLE;
		seq_printf(s, "Bitmap search   : %ld/%ld/%ld (min/avg/max CPU Cycles/IOVA page)\n",
		           min, avg, max);
	}
#endif
#ifndef ALLOW_IOV_BYPASS
	 seq_printf(s, "IOVA bypass disabled\n");
#endif
	return 0;
}

static struct seq_operations ioc_seq_ops = {
	.start = ioc_start,
	.next  = ioc_next,
	.stop  = ioc_stop,
	.show  = ioc_show
};

static int
ioc_open(struct inode *inode, struct file *file)
{
	return seq_open(file, &ioc_seq_ops);
}

static struct file_operations ioc_fops = {
	.open    = ioc_open,
	.read    = seq_read,
	.llseek  = seq_lseek,
	.release = seq_release
};

static void __init
ioc_proc_init(void)
{
	struct proc_dir_entry *dir, *entry;

	dir = proc_mkdir("bus/mckinley", NULL);
	if (!dir)
		return;

	entry = create_proc_entry(ioc_list->name, 0, dir);
	if (entry)
		entry->proc_fops = &ioc_fops;
}
#endif

static void
sba_connect_bus(struct pci_bus *bus)
{
	acpi_handle handle, parent;
	acpi_status status;
	struct ioc *ioc;

	if (!PCI_CONTROLLER(bus))
		panic(PFX "no sysdata on bus %d!\n", bus->number);

	if (PCI_CONTROLLER(bus)->iommu)
		return;

	handle = PCI_CONTROLLER(bus)->acpi_handle;
	if (!handle)
		return;

	/*
	 * The IOC scope encloses PCI root bridges in the ACPI
	 * namespace, so work our way out until we find an IOC we
	 * claimed previously.
	 */
	do {
		for (ioc = ioc_list; ioc; ioc = ioc->next)
			if (ioc->handle == handle) {
				PCI_CONTROLLER(bus)->iommu = ioc;
				return;
			}

		status = acpi_get_parent(handle, &parent);
		handle = parent;
	} while (ACPI_SUCCESS(status));

	printk(KERN_WARNING "No IOC for PCI Bus %04x:%02x in ACPI\n", pci_domain_nr(bus), bus->number);
}

#ifdef CONFIG_NUMA
static void __init
sba_map_ioc_to_node(struct ioc *ioc, acpi_handle handle)
{
	struct acpi_buffer buffer = {ACPI_ALLOCATE_BUFFER, NULL};
	union acpi_object *obj;
	acpi_handle phandle;
	unsigned int node;

	ioc->node = MAX_NUMNODES;

	/*
	 * Check for a _PXM on this node first.  We don't typically see
	 * one here, so we'll end up getting it from the parent.
	 */
	if (ACPI_FAILURE(acpi_evaluate_object(handle, "_PXM", NULL, &buffer))) {
		if (ACPI_FAILURE(acpi_get_parent(handle, &phandle)))
			return;

		/* Reset the acpi buffer */
		buffer.length = ACPI_ALLOCATE_BUFFER;
		buffer.pointer = NULL;

		if (ACPI_FAILURE(acpi_evaluate_object(phandle, "_PXM", NULL,
		                                      &buffer)))
			return;
	}

	if (!buffer.length || !buffer.pointer)
		return;

	obj = buffer.pointer;

	if (obj->type != ACPI_TYPE_INTEGER ||
	    obj->integer.value >= MAX_PXM_DOMAINS) {
		acpi_os_free(buffer.pointer);
		return;
	}

	node = pxm_to_nid_map[obj->integer.value];
	acpi_os_free(buffer.pointer);

	if (node >= MAX_NUMNODES || !node_online(node))
		return;

	ioc->node = node;
	return;
}
#else
#define sba_map_ioc_to_node(ioc, handle)
#endif

static int __init
acpi_sba_ioc_add(struct acpi_device *device)
{
	struct ioc *ioc;
	acpi_status status;
	u64 hpa, length;
	struct acpi_buffer buffer;
	struct acpi_device_info *dev_info;

	status = hp_acpi_csr_space(device->handle, &hpa, &length);
	if (ACPI_FAILURE(status))
		return 1;

	buffer.length = ACPI_ALLOCATE_LOCAL_BUFFER;
	status = acpi_get_object_info(device->handle, &buffer);
	if (ACPI_FAILURE(status))
		return 1;
	dev_info = buffer.pointer;

	/*
	 * For HWP0001, only SBA appears in ACPI namespace.  It encloses the PCI
	 * root bridges, and its CSR space includes the IOC function.
	 */
	if (strncmp("HWP0001", dev_info->hardware_id.value, 7) == 0) {
		hpa += ZX1_IOC_OFFSET;
		/* zx1 based systems default to kernel page size iommu pages */
		if (!iovp_shift)
			iovp_shift = min(PAGE_SHIFT, 16);
	}
	ACPI_MEM_FREE(dev_info);

	/*
	 * default anything not caught above or specified on cmdline to 4k
	 * iommu page size
	 */
	if (!iovp_shift)
		iovp_shift = 12;

	ioc = ioc_init(hpa, device->handle);
	if (!ioc)
		return 1;

	/* setup NUMA node association */
	sba_map_ioc_to_node(ioc, device->handle);
	return 0;
}

static struct acpi_driver acpi_sba_ioc_driver = {
	.name		= "IOC IOMMU Driver",
	.ids		= "HWP0001,HWP0004",
	.ops		= {
		.add	= acpi_sba_ioc_add,
	},
};

static int __init
sba_init(void)
{
	acpi_bus_register_driver(&acpi_sba_ioc_driver);
	if (!ioc_list)
		return 0;

#ifdef CONFIG_PCI
	{
		struct pci_bus *b = NULL;
		while ((b = pci_find_next_bus(b)) != NULL)
			sba_connect_bus(b);
	}
#endif

#ifdef CONFIG_PROC_FS
	ioc_proc_init();
#endif
	return 0;
}

subsys_initcall(sba_init); /* must be initialized after ACPI etc., but before any drivers... */

extern void dig_setup(char**);
/*
 * MAX_DMA_ADDRESS needs to be setup prior to paging_init to do any good,
 * so we use the platform_setup hook to fix it up.
 */
void __init
sba_setup(char **cmdline_p)
{
	MAX_DMA_ADDRESS = ~0UL;
	dig_setup(cmdline_p);
}

static int __init
nosbagart(char *str)
{
	reserve_sba_gart = 0;
	return 1;
}

int
sba_dma_supported (struct device *dev, u64 mask)
{
	/* make sure it's at least 32bit capable */
	return ((mask & 0xFFFFFFFFUL) == 0xFFFFFFFFUL);
}

int
sba_dma_mapping_error (dma_addr_t dma_addr)
{
	return 0;
}

__setup("nosbagart", nosbagart);

static int __init
sba_page_override(char *str)
{
	unsigned long page_size;

	page_size = memparse(str, &str);
	switch (page_size) {
		case 4096:
		case 8192:
		case 16384:
		case 65536:
			iovp_shift = ffs(page_size) - 1;
			break;
		default:
			printk("%s: unknown/unsupported iommu page size %ld\n",
			       __FUNCTION__, page_size);
	}

	return 1;
}

__setup("sbapagesize=",sba_page_override);

EXPORT_SYMBOL(sba_dma_mapping_error);
EXPORT_SYMBOL(sba_map_single);
EXPORT_SYMBOL(sba_unmap_single);
EXPORT_SYMBOL(sba_map_sg);
EXPORT_SYMBOL(sba_unmap_sg);
EXPORT_SYMBOL(sba_dma_supported);
EXPORT_SYMBOL(sba_alloc_coherent);
EXPORT_SYMBOL(sba_free_coherent);