/**
* sn_dma_map_single_attrs - map a single page for DMA
* @dev: device to map for
* @cpu_addr: kernel virtual address of the region to map
* @size: size of the region
* @direction: DMA direction
* @attrs: optional dma attributes
*
* Map the region pointed to by @cpu_addr for DMA and return the
* DMA address.
*
* We map this to the one step pcibr_dmamap_trans interface rather than
* the two step pcibr_dmamap_alloc/pcibr_dmamap_addr because we have
* no way of saving the dmamap handle from the alloc to later free
* (which is pretty much unacceptable).
*
* mappings with the DMA_ATTR_WRITE_BARRIER get mapped with
* dma_map_consistent() so that writes force a flush of pending DMA.
* (See "SGI Altix Architecture Considerations for Linux Device Drivers",
* Document Number: 007-4763-001)
*
* TODO: simplify our interface;
* figure out how to save dmamap handle so can use two step.
*/
dma_addr_t sn_dma_map_single_attrs(struct device *dev, void *cpu_addr,
size_t size, int direction,
struct dma_attrs *attrs)
{
dma_addr_t dma_addr;
unsigned long phys_addr;
struct pci_dev *pdev = to_pci_dev(dev);
struct sn_pcibus_provider *provider = SN_PCIDEV_BUSPROVIDER(pdev);
int dmabarr;
dmabarr = dma_get_attr(DMA_ATTR_WRITE_BARRIER, attrs);
BUG_ON(dev->bus != &pci_bus_type);
phys_addr = __pa(cpu_addr);
if (dmabarr)
dma_addr = provider->dma_map_consistent(pdev, phys_addr,
size, SN_DMA_ADDR_PHYS);
else
dma_addr = provider->dma_map(pdev, phys_addr, size,
SN_DMA_ADDR_PHYS);
if (!dma_addr) {
printk(KERN_ERR "%s: out of ATEs\n", __func__);
return 0;
}
return dma_addr;
}
/**
* sn_dma_map_sg - map a scatterlist for DMA
* @dev: device to map for
* @sg: scatterlist to map
* @nhwentries: number of entries
* @direction: direction of the DMA transaction
*
* Maps each entry of @sg for DMA.
*/
int sn_dma_map_sg(struct device *dev, struct scatterlist *sg, int nhwentries,
int direction)
{
unsigned long phys_addr;
struct scatterlist *saved_sg = sg;
struct pci_dev *pdev = to_pci_dev(dev);
struct sn_pcibus_provider *provider = SN_PCIDEV_BUSPROVIDER(pdev);
int i;
BUG_ON(dev->bus != &pci_bus_type);
/*
* Setup a DMA address for each entry in the scatterlist.
*/
for (i = 0; i < nhwentries; i++, sg++) {
phys_addr = SG_ENT_PHYS_ADDRESS(sg);
sg->dma_address = provider->dma_map(pdev,
phys_addr, sg->length);
if (!sg->dma_address) {
printk(KERN_ERR "%s: out of ATEs\n", __FUNCTION__);
/*
* Free any successfully allocated entries.
*/
if (i > 0)
sn_dma_unmap_sg(dev, saved_sg, i, direction);
return 0;
}
sg->dma_length = sg->length;
}
return nhwentries;
}
void sn_teardown_msi_irq(unsigned int irq)
{
nasid_t nasid;
int widget;
struct pci_dev *pdev;
struct pcidev_info *sn_pdev;
struct sn_irq_info *sn_irq_info;
struct pcibus_bussoft *bussoft;
struct sn_pcibus_provider *provider;
sn_irq_info = sn_msi_info[irq].sn_irq_info;
if (sn_irq_info == NULL || sn_irq_info->irq_int_bit >= 0)
return;
sn_pdev = (struct pcidev_info *)sn_irq_info->irq_pciioinfo;
pdev = sn_pdev->pdi_linux_pcidev;
provider = SN_PCIDEV_BUSPROVIDER(pdev);
(*provider->dma_unmap)(pdev,
sn_msi_info[irq].pci_addr,
PCI_DMA_FROMDEVICE);
sn_msi_info[irq].pci_addr = 0;
bussoft = SN_PCIDEV_BUSSOFT(pdev);
nasid = NASID_GET(bussoft->bs_base);
widget = (nasid & 1) ?
TIO_SWIN_WIDGETNUM(bussoft->bs_base) :
SWIN_WIDGETNUM(bussoft->bs_base);
sn_intr_free(nasid, widget, sn_irq_info);
sn_msi_info[irq].sn_irq_info = NULL;
destroy_irq(irq);
}
static int sn_set_msi_irq_affinity(struct irq_data *data,
const struct cpumask *cpu_mask, bool force)
{
struct msi_msg msg;
int slice;
nasid_t nasid;
u64 bus_addr;
struct pci_dev *pdev;
struct pcidev_info *sn_pdev;
struct sn_irq_info *sn_irq_info;
struct sn_irq_info *new_irq_info;
struct sn_pcibus_provider *provider;
unsigned int cpu, irq = data->irq;
cpu = cpumask_first_and(cpu_mask, cpu_online_mask);
sn_irq_info = sn_msi_info[irq].sn_irq_info;
if (sn_irq_info == NULL || sn_irq_info->irq_int_bit >= 0)
return -1;
/*
* Release XIO resources for the old MSI PCI address
*/
__get_cached_msi_msg(data->msi_desc, &msg);
sn_pdev = (struct pcidev_info *)sn_irq_info->irq_pciioinfo;
pdev = sn_pdev->pdi_linux_pcidev;
provider = SN_PCIDEV_BUSPROVIDER(pdev);
bus_addr = (u64)(msg.address_hi) << 32 | (u64)(msg.address_lo);
(*provider->dma_unmap)(pdev, bus_addr, PCI_DMA_FROMDEVICE);
sn_msi_info[irq].pci_addr = 0;
nasid = cpuid_to_nasid(cpu);
slice = cpuid_to_slice(cpu);
new_irq_info = sn_retarget_vector(sn_irq_info, nasid, slice);
sn_msi_info[irq].sn_irq_info = new_irq_info;
if (new_irq_info == NULL)
return -1;
/*
* Map the xio address into bus space
*/
bus_addr = (*provider->dma_map_consistent)(pdev,
new_irq_info->irq_xtalkaddr,
sizeof(new_irq_info->irq_xtalkaddr),
SN_DMA_MSI|SN_DMA_ADDR_XIO);
sn_msi_info[irq].pci_addr = bus_addr;
msg.address_hi = (u32)(bus_addr >> 32);
msg.address_lo = (u32)(bus_addr & 0x00000000ffffffff);
pci_write_msi_msg(irq, &msg);
cpumask_copy(data->affinity, cpu_mask);
return 0;
}
/**
* sn_dma_unmap_single - unamp a DMA mapped page
* @dev: device to sync
* @dma_addr: DMA address to sync
* @size: size of region
* @direction: DMA direction
*
* This routine is supposed to sync the DMA region specified
* by @dma_handle into the coherence domain. On SN, we're always cache
* coherent, so we just need to free any ATEs associated with this mapping.
*/
void sn_dma_unmap_single(struct device *dev, dma_addr_t dma_addr, size_t size,
int direction)
{
struct pci_dev *pdev = to_pci_dev(dev);
struct sn_pcibus_provider *provider = SN_PCIDEV_BUSPROVIDER(pdev);
BUG_ON(dev->bus != &pci_bus_type);
provider->dma_unmap(pdev, dma_addr, direction);
}
/**
* sn_pci_free_coherent - free memory associated with coherent DMAable region
* @dev: device to free for
* @size: size to free
* @cpu_addr: kernel virtual address to free
* @dma_handle: DMA address associated with this region
*
* Frees the memory allocated by dma_alloc_coherent(), potentially unmapping
* any associated IOMMU mappings.
*/
void sn_dma_free_coherent(struct device *dev, size_t size, void *cpu_addr,
dma_addr_t dma_handle)
{
struct pci_dev *pdev = to_pci_dev(dev);
struct sn_pcibus_provider *provider = SN_PCIDEV_BUSPROVIDER(pdev);
BUG_ON(dev->bus != &pci_bus_type);
provider->dma_unmap(pdev, dma_handle, 0);
free_pages((unsigned long)cpu_addr, get_order(size));
}
/**
* sn_dma_unmap_single_attrs - unamp a DMA mapped page
* @dev: device to sync
* @dma_addr: DMA address to sync
* @size: size of region
* @direction: DMA direction
* @attrs: optional dma attributes
*
* This routine is supposed to sync the DMA region specified
* by @dma_handle into the coherence domain. On SN, we're always cache
* coherent, so we just need to free any ATEs associated with this mapping.
*/
static void sn_dma_unmap_page(struct device *dev, dma_addr_t dma_addr,
size_t size, enum dma_data_direction dir,
struct dma_attrs *attrs)
{
struct pci_dev *pdev = to_pci_dev(dev);
struct sn_pcibus_provider *provider = SN_PCIDEV_BUSPROVIDER(pdev);
BUG_ON(dev->bus != &pci_bus_type);
provider->dma_unmap(pdev, dma_addr, dir);
}
static void
sn_msi_target(unsigned int vector, unsigned int cpu,
u32 *addr_hi, u32 *addr_lo)
{
int slice;
nasid_t nasid;
u64 bus_addr;
struct pci_dev *pdev;
struct pcidev_info *sn_pdev;
struct sn_irq_info *sn_irq_info;
struct sn_irq_info *new_irq_info;
struct sn_pcibus_provider *provider;
sn_irq_info = sn_msi_info[vector].sn_irq_info;
if (sn_irq_info == NULL || sn_irq_info->irq_int_bit >= 0)
return;
/*
* Release XIO resources for the old MSI PCI address
*/
sn_pdev = (struct pcidev_info *)sn_irq_info->irq_pciioinfo;
pdev = sn_pdev->pdi_linux_pcidev;
provider = SN_PCIDEV_BUSPROVIDER(pdev);
bus_addr = (u64)(*addr_hi) << 32 | (u64)(*addr_lo);
(*provider->dma_unmap)(pdev, bus_addr, PCI_DMA_FROMDEVICE);
sn_msi_info[vector].pci_addr = 0;
nasid = cpuid_to_nasid(cpu);
slice = cpuid_to_slice(cpu);
new_irq_info = sn_retarget_vector(sn_irq_info, nasid, slice);
sn_msi_info[vector].sn_irq_info = new_irq_info;
if (new_irq_info == NULL)
return;
/*
* Map the xio address into bus space
*/
bus_addr = (*provider->dma_map_consistent)(pdev,
new_irq_info->irq_xtalkaddr,
sizeof(new_irq_info->irq_xtalkaddr),
SN_DMA_MSI|SN_DMA_ADDR_XIO);
sn_msi_info[vector].pci_addr = bus_addr;
*addr_hi = (u32)(bus_addr >> 32);
*addr_lo = (u32)(bus_addr & 0x00000000ffffffff);
}
/**
* sn_dma_alloc_coherent - allocate memory for coherent DMA
* @dev: device to allocate for
* @size: size of the region
* @dma_handle: DMA (bus) address
* @flags: memory allocation flags
*
* dma_alloc_coherent() returns a pointer to a memory region suitable for
* coherent DMA traffic to/from a PCI device. On SN platforms, this means
* that @dma_handle will have the %PCIIO_DMA_CMD flag set.
*
* This interface is usually used for "command" streams (e.g. the command
* queue for a SCSI controller). See Documentation/DMA-API.txt for
* more information.
*/
void *sn_dma_alloc_coherent(struct device *dev, size_t size,
dma_addr_t * dma_handle, gfp_t flags)
{
void *cpuaddr;
unsigned long phys_addr;
int node;
struct pci_dev *pdev = to_pci_dev(dev);
struct sn_pcibus_provider *provider = SN_PCIDEV_BUSPROVIDER(pdev);
BUG_ON(dev->bus != &pci_bus_type);
/*
* Allocate the memory.
*/
node = pcibus_to_node(pdev->bus);
if (likely(node >=0)) {
struct page *p = alloc_pages_node(node, flags, get_order(size));
if (likely(p))
cpuaddr = page_address(p);
else
return NULL;
} else
cpuaddr = (void *)__get_free_pages(flags, get_order(size));
if (unlikely(!cpuaddr))
return NULL;
memset(cpuaddr, 0x0, size);
/* physical addr. of the memory we just got */
phys_addr = __pa(cpuaddr);
/*
* 64 bit address translations should never fail.
* 32 bit translations can fail if there are insufficient mapping
* resources.
*/
*dma_handle = provider->dma_map_consistent(pdev, phys_addr, size,
SN_DMA_ADDR_PHYS);
if (!*dma_handle) {
printk(KERN_ERR "%s: out of ATEs\n", __func__);
free_pages((unsigned long)cpuaddr, get_order(size));
return NULL;
}
return cpuaddr;
}
/**
* sn_dma_unmap_sg - unmap a DMA scatterlist
* @dev: device to unmap
* @sg: scatterlist to unmap
* @nhwentries: number of scatterlist entries
* @direction: DMA direction
*
* Unmap a set of streaming mode DMA translations.
*/
void sn_dma_unmap_sg(struct device *dev, struct scatterlist *sg,
int nhwentries, int direction)
{
int i;
struct pci_dev *pdev = to_pci_dev(dev);
struct sn_pcibus_provider *provider = SN_PCIDEV_BUSPROVIDER(pdev);
BUG_ON(dev->bus != &pci_bus_type);
for (i = 0; i < nhwentries; i++, sg++) {
provider->dma_unmap(pdev, sg->dma_address, direction);
sg->dma_address = (dma_addr_t) NULL;
sg->dma_length = 0;
}
}
/**
* sn_dma_unmap_sg_attrs - unmap a DMA scatterlist
* @dev: device to unmap
* @sg: scatterlist to unmap
* @nhwentries: number of scatterlist entries
* @direction: DMA direction
* @attrs: optional dma attributes
*
* Unmap a set of streaming mode DMA translations.
*/
void sn_dma_unmap_sg_attrs(struct device *dev, struct scatterlist *sgl,
int nhwentries, int direction,
struct dma_attrs *attrs)
{
int i;
struct pci_dev *pdev = to_pci_dev(dev);
struct sn_pcibus_provider *provider = SN_PCIDEV_BUSPROVIDER(pdev);
struct scatterlist *sg;
BUG_ON(dev->bus != &pci_bus_type);
for_each_sg(sgl, sg, nhwentries, i) {
provider->dma_unmap(pdev, sg->dma_address, direction);
sg->dma_address = (dma_addr_t) NULL;
sg->dma_length = 0;
}
/**
* sn_dma_map_single - map a single page for DMA
* @dev: device to map for
* @cpu_addr: kernel virtual address of the region to map
* @size: size of the region
* @direction: DMA direction
*
* Map the region pointed to by @cpu_addr for DMA and return the
* DMA address.
*
* We map this to the one step pcibr_dmamap_trans interface rather than
* the two step pcibr_dmamap_alloc/pcibr_dmamap_addr because we have
* no way of saving the dmamap handle from the alloc to later free
* (which is pretty much unacceptable).
*
* TODO: simplify our interface;
* figure out how to save dmamap handle so can use two step.
*/
dma_addr_t sn_dma_map_single(struct device *dev, void *cpu_addr, size_t size,
int direction)
{
dma_addr_t dma_addr;
unsigned long phys_addr;
struct pci_dev *pdev = to_pci_dev(dev);
struct sn_pcibus_provider *provider = SN_PCIDEV_BUSPROVIDER(pdev);
BUG_ON(dev->bus != &pci_bus_type);
phys_addr = __pa(cpu_addr);
dma_addr = provider->dma_map(pdev, phys_addr, size);
if (!dma_addr) {
printk(KERN_ERR "%s: out of ATEs\n", __FUNCTION__);
return 0;
}
return dma_addr;
}
int sn_setup_msi_irq(struct pci_dev *pdev, struct msi_desc *entry)
{
struct msi_msg msg;
int widget;
int status;
nasid_t nasid;
u64 bus_addr;
struct sn_irq_info *sn_irq_info;
struct pcibus_bussoft *bussoft = SN_PCIDEV_BUSSOFT(pdev);
struct sn_pcibus_provider *provider = SN_PCIDEV_BUSPROVIDER(pdev);
int irq;
if (!entry->msi_attrib.is_64)
return -EINVAL;
if (bussoft == NULL)
return -EINVAL;
if (provider == NULL || provider->dma_map_consistent == NULL)
return -EINVAL;
irq = create_irq();
if (irq < 0)
return irq;
/*
* Set up the vector plumbing. Let the prom (via sn_intr_alloc)
* decide which cpu to direct this msi at by default.
*/
nasid = NASID_GET(bussoft->bs_base);
widget = (nasid & 1) ?
TIO_SWIN_WIDGETNUM(bussoft->bs_base) :
SWIN_WIDGETNUM(bussoft->bs_base);
sn_irq_info = kzalloc(sizeof(struct sn_irq_info), GFP_KERNEL);
if (! sn_irq_info) {
destroy_irq(irq);
return -ENOMEM;
}
status = sn_intr_alloc(nasid, widget, sn_irq_info, irq, -1, -1);
if (status) {
kfree(sn_irq_info);
destroy_irq(irq);
return -ENOMEM;
}
sn_irq_info->irq_int_bit = -1; /* mark this as an MSI irq */
sn_irq_fixup(pdev, sn_irq_info);
/* Prom probably should fill these in, but doesn't ... */
sn_irq_info->irq_bridge_type = bussoft->bs_asic_type;
sn_irq_info->irq_bridge = (void *)bussoft->bs_base;
/*
* Map the xio address into bus space
*/
bus_addr = (*provider->dma_map_consistent)(pdev,
sn_irq_info->irq_xtalkaddr,
sizeof(sn_irq_info->irq_xtalkaddr),
SN_DMA_MSI|SN_DMA_ADDR_XIO);
if (! bus_addr) {
sn_intr_free(nasid, widget, sn_irq_info);
kfree(sn_irq_info);
destroy_irq(irq);
return -ENOMEM;
}
sn_msi_info[irq].sn_irq_info = sn_irq_info;
sn_msi_info[irq].pci_addr = bus_addr;
msg.address_hi = (u32)(bus_addr >> 32);
msg.address_lo = (u32)(bus_addr & 0x00000000ffffffff);
/*
* In the SN platform, bit 16 is a "send vector" bit which
* must be present in order to move the vector through the system.
*/
msg.data = 0x100 + irq;
irq_set_msi_desc(irq, entry);
pci_write_msi_msg(irq, &msg);
irq_set_chip_and_handler(irq, &sn_msi_chip, handle_edge_irq);
return 0;
}
int sn_setup_msi_irq(struct pci_dev *pdev, struct msi_desc *entry)
{
struct msi_msg msg;
int widget;
int status;
nasid_t nasid;
u64 bus_addr;
struct sn_irq_info *sn_irq_info;
struct pcibus_bussoft *bussoft = SN_PCIDEV_BUSSOFT(pdev);
struct sn_pcibus_provider *provider = SN_PCIDEV_BUSPROVIDER(pdev);
int irq;
if (!entry->msi_attrib.is_64)
return -EINVAL;
if (bussoft == NULL)
return -EINVAL;
if (provider == NULL || provider->dma_map_consistent == NULL)
return -EINVAL;
irq = create_irq();
if (irq < 0)
return irq;
nasid = NASID_GET(bussoft->bs_base);
widget = (nasid & 1) ?
TIO_SWIN_WIDGETNUM(bussoft->bs_base) :
SWIN_WIDGETNUM(bussoft->bs_base);
sn_irq_info = kzalloc(sizeof(struct sn_irq_info), GFP_KERNEL);
if (! sn_irq_info) {
destroy_irq(irq);
return -ENOMEM;
}
status = sn_intr_alloc(nasid, widget, sn_irq_info, irq, -1, -1);
if (status) {
kfree(sn_irq_info);
destroy_irq(irq);
return -ENOMEM;
}
sn_irq_info->irq_int_bit = -1;
sn_irq_fixup(pdev, sn_irq_info);
sn_irq_info->irq_bridge_type = bussoft->bs_asic_type;
sn_irq_info->irq_bridge = (void *)bussoft->bs_base;
bus_addr = (*provider->dma_map_consistent)(pdev,
sn_irq_info->irq_xtalkaddr,
sizeof(sn_irq_info->irq_xtalkaddr),
SN_DMA_MSI|SN_DMA_ADDR_XIO);
if (! bus_addr) {
sn_intr_free(nasid, widget, sn_irq_info);
kfree(sn_irq_info);
destroy_irq(irq);
return -ENOMEM;
}
sn_msi_info[irq].sn_irq_info = sn_irq_info;
sn_msi_info[irq].pci_addr = bus_addr;
msg.address_hi = (u32)(bus_addr >> 32);
msg.address_lo = (u32)(bus_addr & 0x00000000ffffffff);
msg.data = 0x100 + irq;
irq_set_msi_desc(irq, entry);
write_msi_msg(irq, &msg);
irq_set_chip_and_handler(irq, &sn_msi_chip, handle_edge_irq);
return 0;
}
int
sn_msi_setup(struct pci_dev *pdev, unsigned int vector,
u32 *addr_hi, u32 *addr_lo, u32 *data)
{
int widget;
int status;
nasid_t nasid;
u64 bus_addr;
struct sn_irq_info *sn_irq_info;
struct pcibus_bussoft *bussoft = SN_PCIDEV_BUSSOFT(pdev);
struct sn_pcibus_provider *provider = SN_PCIDEV_BUSPROVIDER(pdev);
if (bussoft == NULL)
return -EINVAL;
if (provider == NULL || provider->dma_map_consistent == NULL)
return -EINVAL;
/*
* Set up the vector plumbing. Let the prom (via sn_intr_alloc)
* decide which cpu to direct this msi at by default.
*/
nasid = NASID_GET(bussoft->bs_base);
widget = (nasid & 1) ?
TIO_SWIN_WIDGETNUM(bussoft->bs_base) :
SWIN_WIDGETNUM(bussoft->bs_base);
sn_irq_info = kzalloc(sizeof(struct sn_irq_info), GFP_KERNEL);
if (! sn_irq_info)
return -ENOMEM;
status = sn_intr_alloc(nasid, widget, sn_irq_info, vector, -1, -1);
if (status) {
kfree(sn_irq_info);
return -ENOMEM;
}
sn_irq_info->irq_int_bit = -1; /* mark this as an MSI irq */
sn_irq_fixup(pdev, sn_irq_info);
/* Prom probably should fill these in, but doesn't ... */
sn_irq_info->irq_bridge_type = bussoft->bs_asic_type;
sn_irq_info->irq_bridge = (void *)bussoft->bs_base;
/*
* Map the xio address into bus space
*/
bus_addr = (*provider->dma_map_consistent)(pdev,
sn_irq_info->irq_xtalkaddr,
sizeof(sn_irq_info->irq_xtalkaddr),
SN_DMA_MSI|SN_DMA_ADDR_XIO);
if (! bus_addr) {
sn_intr_free(nasid, widget, sn_irq_info);
kfree(sn_irq_info);
return -ENOMEM;
}
sn_msi_info[vector].sn_irq_info = sn_irq_info;
sn_msi_info[vector].pci_addr = bus_addr;
*addr_hi = (u32)(bus_addr >> 32);
*addr_lo = (u32)(bus_addr & 0x00000000ffffffff);
/*
* In the SN platform, bit 16 is a "send vector" bit which
* must be present in order to move the vector through the system.
*/
*data = 0x100 + (unsigned int)vector;
#ifdef CONFIG_SMP
set_irq_affinity_info((vector & 0xff), sn_irq_info->irq_cpuid, 0);
#endif
return 0;
}