#include <linux/mm.h>
#include <linux/string.h>
#include <linux/pci.h>
+#include <linux/scatterlist.h>
+#include <linux/iommu-helper.h>
#include <asm/byteorder.h>
#include <asm/io.h>
#include <linux/proc_fs.h>
#include <linux/seq_file.h>
+#include <asm/ropes.h>
#include <asm/mckinley.h> /* for proc_mckinley_root */
#include <asm/runway.h> /* for proc_runway_root */
#include <asm/pdc.h> /* for PDC_MODEL_* */
#define MODULE_NAME "SBA"
-#ifdef CONFIG_PROC_FS
-/* depends on proc fs support. But costs CPU performance */
-#undef SBA_COLLECT_STATS
-#endif
-
/*
** The number of debug flags is a clue - this code is fragile.
** Don't even think about messing with it unless you have
#define DBG_RES(x...)
#endif
-#if defined(CONFIG_64BIT)
-/* "low end" PA8800 machines use ZX1 chipset: PAT PDC and only run 64-bit */
-#define ZX1_SUPPORT
-#endif
-
#define SBA_INLINE __inline__
-
-/*
-** The number of pdir entries to "free" before issueing
-** a read to PCOM register to flush out PCOM writes.
-** Interacts with allocation granularity (ie 4 or 8 entries
-** allocated and free'd/purged at a time might make this
-** less interesting).
-*/
-#define DELAYED_RESOURCE_CNT 16
-
#define DEFAULT_DMA_HINT_REG 0
-#define ASTRO_RUNWAY_PORT 0x582
-#define IKE_MERCED_PORT 0x803
-#define REO_MERCED_PORT 0x804
-#define REOG_MERCED_PORT 0x805
-#define PLUTO_MCKINLEY_PORT 0x880
-
-#define SBA_FUNC_ID 0x0000 /* function id */
-#define SBA_FCLASS 0x0008 /* function class, bist, header, rev... */
-
-static inline int IS_ASTRO(struct parisc_device *d) {
- return d->id.hversion == ASTRO_RUNWAY_PORT;
-}
-
-static inline int IS_IKE(struct parisc_device *d) {
- return d->id.hversion == IKE_MERCED_PORT;
-}
-
-static inline int IS_PLUTO(struct parisc_device *d) {
- return d->id.hversion == PLUTO_MCKINLEY_PORT;
-}
-
-#define SBA_FUNC_SIZE 4096 /* SBA configuration function reg set */
-
-#define ASTRO_IOC_OFFSET (32 * SBA_FUNC_SIZE)
-#define PLUTO_IOC_OFFSET (1 * SBA_FUNC_SIZE)
-/* Ike's IOC's occupy functions 2 and 3 */
-#define IKE_IOC_OFFSET(p) ((p+2) * SBA_FUNC_SIZE)
-
-#define IOC_CTRL 0x8 /* IOC_CTRL offset */
-#define IOC_CTRL_TC (1 << 0) /* TOC Enable */
-#define IOC_CTRL_CE (1 << 1) /* Coalesce Enable */
-#define IOC_CTRL_DE (1 << 2) /* Dillon Enable */
-#define IOC_CTRL_RM (1 << 8) /* Real Mode */
-#define IOC_CTRL_NC (1 << 9) /* Non Coherent Mode */
-#define IOC_CTRL_D4 (1 << 11) /* Disable 4-byte coalescing */
-#define IOC_CTRL_DD (1 << 13) /* Disable distr. LMMIO range coalescing */
-
-#define MAX_IOC 2 /* per Ike. Pluto/Astro only have 1. */
-
-#define ROPES_PER_IOC 8 /* per Ike half or Pluto/Astro */
-
-
-/*
-** Offsets into MBIB (Function 0 on Ike and hopefully Astro)
-** Firmware programs this stuff. Don't touch it.
-*/
-#define LMMIO_DIRECT0_BASE 0x300
-#define LMMIO_DIRECT0_MASK 0x308
-#define LMMIO_DIRECT0_ROUTE 0x310
-
-#define LMMIO_DIST_BASE 0x360
-#define LMMIO_DIST_MASK 0x368
-#define LMMIO_DIST_ROUTE 0x370
-
-#define IOS_DIST_BASE 0x390
-#define IOS_DIST_MASK 0x398
-#define IOS_DIST_ROUTE 0x3A0
-
-#define IOS_DIRECT_BASE 0x3C0
-#define IOS_DIRECT_MASK 0x3C8
-#define IOS_DIRECT_ROUTE 0x3D0
-
-/*
-** Offsets into I/O TLB (Function 2 and 3 on Ike)
-*/
-#define ROPE0_CTL 0x200 /* "regbus pci0" */
-#define ROPE1_CTL 0x208
-#define ROPE2_CTL 0x210
-#define ROPE3_CTL 0x218
-#define ROPE4_CTL 0x220
-#define ROPE5_CTL 0x228
-#define ROPE6_CTL 0x230
-#define ROPE7_CTL 0x238
-
-#define IOC_ROPE0_CFG 0x500 /* pluto only */
-#define IOC_ROPE_AO 0x10 /* Allow "Relaxed Ordering" */
-
-
-
-#define HF_ENABLE 0x40
-
-
-#define IOC_IBASE 0x300 /* IO TLB */
-#define IOC_IMASK 0x308
-#define IOC_PCOM 0x310
-#define IOC_TCNFG 0x318
-#define IOC_PDIR_BASE 0x320
-
-/* AGP GART driver looks for this */
-#define SBA_IOMMU_COOKIE 0x0000badbadc0ffeeUL
-
-
-/*
-** IOC supports 4/8/16/64KB page sizes (see TCNFG register)
-** It's safer (avoid memory corruption) to keep DMA page mappings
-** equivalently sized to VM PAGE_SIZE.
-**
-** We really can't avoid generating a new mapping for each
-** page since the Virtual Coherence Index has to be generated
-** and updated for each page.
-**
-** PAGE_SIZE could be greater than IOVP_SIZE. But not the inverse.
-*/
-#define IOVP_SIZE PAGE_SIZE
-#define IOVP_SHIFT PAGE_SHIFT
-#define IOVP_MASK PAGE_MASK
-
-#define SBA_PERF_CFG 0x708 /* Performance Counter stuff */
-#define SBA_PERF_MASK1 0x718
-#define SBA_PERF_MASK2 0x730
-
-
-/*
-** Offsets into PCI Performance Counters (functions 12 and 13)
-** Controlled by PERF registers in function 2 & 3 respectively.
-*/
-#define SBA_PERF_CNT1 0x200
-#define SBA_PERF_CNT2 0x208
-#define SBA_PERF_CNT3 0x210
-
-
-struct ioc {
- void __iomem *ioc_hpa; /* I/O MMU base address */
- char *res_map; /* resource map, bit == pdir entry */
- u64 *pdir_base; /* physical base address */
- unsigned long ibase; /* pdir IOV Space base - shared w/lba_pci */
- unsigned long imask; /* pdir IOV Space mask - shared w/lba_pci */
-#ifdef ZX1_SUPPORT
- unsigned long iovp_mask; /* help convert IOVA to IOVP */
-#endif
- unsigned long *res_hint; /* next avail IOVP - circular search */
- spinlock_t res_lock;
- unsigned int res_bitshift; /* from the LEFT! */
- unsigned int res_size; /* size of resource map in bytes */
-#ifdef SBA_HINT_SUPPORT
-/* FIXME : DMA HINTs not used */
- unsigned long hint_mask_pdir; /* bits used for DMA hints */
- unsigned int hint_shift_pdir;
-#endif
-#if DELAYED_RESOURCE_CNT > 0
- int saved_cnt;
- struct sba_dma_pair {
- dma_addr_t iova;
- size_t size;
- } saved[DELAYED_RESOURCE_CNT];
-#endif
-
-#ifdef SBA_COLLECT_STATS
-#define SBA_SEARCH_SAMPLE 0x100
- unsigned long avg_search[SBA_SEARCH_SAMPLE];
- unsigned long avg_idx; /* current index into avg_search */
- unsigned long used_pages;
- unsigned long msingle_calls;
- unsigned long msingle_pages;
- unsigned long msg_calls;
- unsigned long msg_pages;
- unsigned long usingle_calls;
- unsigned long usingle_pages;
- unsigned long usg_calls;
- unsigned long usg_pages;
-#endif
-
- /* STUFF We don't need in performance path */
- unsigned int pdir_size; /* in bytes, determined by IOV Space size */
-};
-
-struct sba_device {
- struct sba_device *next; /* list of SBA's in system */
- struct parisc_device *dev; /* dev found in bus walk */
- const char *name;
- void __iomem *sba_hpa; /* base address */
- spinlock_t sba_lock;
- unsigned int flags; /* state/functionality enabled */
- unsigned int hw_rev; /* HW revision of chip */
-
- struct resource chip_resv; /* MMIO reserved for chip */
- struct resource iommu_resv; /* MMIO reserved for iommu */
-
- unsigned int num_ioc; /* number of on-board IOC's */
- struct ioc ioc[MAX_IOC];
-};
-
-
-static struct sba_device *sba_list;
+struct sba_device *sba_list;
+EXPORT_SYMBOL_GPL(sba_list);
static unsigned long ioc_needs_fdc = 0;
/* Looks nice and keeps the compiler happy */
#define SBA_DEV(d) ((struct sba_device *) (d))
+#ifdef CONFIG_AGP_PARISC
+#define SBA_AGP_SUPPORT
+#endif /*CONFIG_AGP_PARISC*/
+
#ifdef SBA_AGP_SUPPORT
-static int reserve_sba_gart = 1;
+static int sba_reserve_agpgart = 1;
+module_param(sba_reserve_agpgart, int, 0444);
+MODULE_PARM_DESC(sba_reserve_agpgart, "Reserve half of IO pdir as AGPGART");
#endif
-#define ROUNDUP(x,y) ((x + ((y)-1)) & ~((y)-1))
-
/************************************
** SBA register read and write support
#define RESMAP_MASK(n) (~0UL << (BITS_PER_LONG - (n)))
#define RESMAP_IDX_MASK (sizeof(unsigned long) - 1)
+static unsigned long ptr_to_pide(struct ioc *ioc, unsigned long *res_ptr,
+ unsigned int bitshiftcnt)
+{
+ return (((unsigned long)res_ptr - (unsigned long)ioc->res_map) << 3)
+ + bitshiftcnt;
+}
/**
* sba_search_bitmap - find free space in IO PDIR resource bitmap
* Cool perf optimization: search for log2(size) bits at a time.
*/
static SBA_INLINE unsigned long
-sba_search_bitmap(struct ioc *ioc, unsigned long bits_wanted)
+sba_search_bitmap(struct ioc *ioc, struct device *dev,
+ unsigned long bits_wanted)
{
unsigned long *res_ptr = ioc->res_hint;
unsigned long *res_end = (unsigned long *) &(ioc->res_map[ioc->res_size]);
- unsigned long pide = ~0UL;
+ unsigned long pide = ~0UL, tpide;
+ unsigned long boundary_size;
+ unsigned long shift;
+ int ret;
+
+ boundary_size = ALIGN((unsigned long long)dma_get_seg_boundary(dev) + 1,
+ 1ULL << IOVP_SHIFT) >> IOVP_SHIFT;
+
+#if defined(ZX1_SUPPORT)
+ BUG_ON(ioc->ibase & ~IOVP_MASK);
+ shift = ioc->ibase >> IOVP_SHIFT;
+#else
+ shift = 0;
+#endif
if (bits_wanted > (BITS_PER_LONG/2)) {
/* Search word at a time - no mask needed */
for(; res_ptr < res_end; ++res_ptr) {
- if (*res_ptr == 0) {
+ tpide = ptr_to_pide(ioc, res_ptr, 0);
+ ret = iommu_is_span_boundary(tpide, bits_wanted,
+ shift,
+ boundary_size);
+ if ((*res_ptr == 0) && !ret) {
*res_ptr = RESMAP_MASK(bits_wanted);
- pide = ((unsigned long)res_ptr - (unsigned long)ioc->res_map);
- pide <<= 3; /* convert to bit address */
+ pide = tpide;
break;
}
}
** SBA HW features in the unmap path.
*/
unsigned long o = 1 << get_order(bits_wanted << PAGE_SHIFT);
- uint bitshiftcnt = ROUNDUP(ioc->res_bitshift, o);
+ uint bitshiftcnt = ALIGN(ioc->res_bitshift, o);
unsigned long mask;
if (bitshiftcnt >= BITS_PER_LONG) {
{
DBG_RES(" %p %lx %lx\n", res_ptr, mask, *res_ptr);
WARN_ON(mask == 0);
- if(((*res_ptr) & mask) == 0) {
+ tpide = ptr_to_pide(ioc, res_ptr, bitshiftcnt);
+ ret = iommu_is_span_boundary(tpide, bits_wanted,
+ shift,
+ boundary_size);
+ if ((((*res_ptr) & mask) == 0) && !ret) {
*res_ptr |= mask; /* mark resources busy! */
- pide = ((unsigned long)res_ptr - (unsigned long)ioc->res_map);
- pide <<= 3; /* convert to bit address */
- pide += bitshiftcnt;
+ pide = tpide;
break;
}
mask >>= o;
* resource bit map.
*/
static int
-sba_alloc_range(struct ioc *ioc, size_t size)
+sba_alloc_range(struct ioc *ioc, struct device *dev, size_t size)
{
unsigned int pages_needed = size >> IOVP_SHIFT;
#ifdef SBA_COLLECT_STATS
#endif
unsigned long pide;
- pide = sba_search_bitmap(ioc, pages_needed);
+ pide = sba_search_bitmap(ioc, dev, pages_needed);
if (pide >= (ioc->res_size << 3)) {
- pide = sba_search_bitmap(ioc, pages_needed);
+ pide = sba_search_bitmap(ioc, dev, pages_needed);
if (pide >= (ioc->res_size << 3))
panic("%s: I/O MMU @ %p is out of mapping resources\n",
__FILE__, ioc->ioc_hpa);
asm("lci 0(%%sr1, %1), %0" : "=r" (ci) : "r" (vba));
pa |= (ci >> 12) & 0xff; /* move CI (8 bits) into lowest byte */
- pa |= 0x8000000000000000ULL; /* set "valid" bit */
+ pa |= SBA_PDIR_VALID_BIT; /* set "valid" bit */
*pdir_ptr = cpu_to_le64(pa); /* swap and store into I/O Pdir */
/*
ioc->msingle_calls++;
ioc->msingle_pages += size >> IOVP_SHIFT;
#endif
- pide = sba_alloc_range(ioc, size);
+ pide = sba_alloc_range(ioc, dev, size);
iovp = (dma_addr_t) pide << IOVP_SHIFT;
DBG_RUN("%s() 0x%p -> 0x%lx\n",
offset = iova & ~IOVP_MASK;
iova ^= offset; /* clear offset bits */
size += offset;
- size = ROUNDUP(size, IOVP_SIZE);
+ size = ALIGN(size, IOVP_SIZE);
spin_lock_irqsave(&ioc->res_lock, flags);
if (!hwdev) {
/* only support PCI */
*dma_handle = 0;
- return 0;
+ return NULL;
}
ret = (void *) __get_free_pages(gfp, get_order(size));
** w/o this association, we wouldn't have coherent DMA!
** Access to the virtual address is what forces a two pass algorithm.
*/
- coalesced = iommu_coalesce_chunks(ioc, sglist, nents, sba_alloc_range);
+ coalesced = iommu_coalesce_chunks(ioc, dev, sglist, nents, sba_alloc_range);
/*
** Program the I/O Pdir
WRITE_REG(ioc->ibase | 31, ioc->ioc_hpa + IOC_PCOM);
#ifdef SBA_AGP_SUPPORT
+{
+ struct klist_iter i;
+ struct device *dev = NULL;
+
/*
** If an AGP device is present, only use half of the IOV space
** for PCI DMA. Unfortunately we can't know ahead of time
** We program the next pdir index after we stop w/ a key for
** the GART code to handshake on.
*/
- device=NULL;
- for (lba = sba->child; lba; lba = lba->sibling) {
+ klist_iter_init(&sba->dev.klist_children, &i);
+ while ((dev = next_device(&i))) {
+ struct parisc_device *lba = to_parisc_device(dev);
if (IS_QUICKSILVER(lba))
- break;
+ agp_found = 1;
}
+ klist_iter_exit(&i);
- if (lba) {
- DBG_INIT("%s: Reserving half of IOVA space for AGP GART support\n", __FUNCTION__);
+ if (agp_found && sba_reserve_agpgart) {
+ printk(KERN_INFO "%s: reserving %dMb of IOVA space for agpgart\n",
+ __FUNCTION__, (iova_space_size/2) >> 20);
ioc->pdir_size /= 2;
- ((u64 *)ioc->pdir_base)[PDIR_INDEX(iova_space_size/2)] = SBA_IOMMU_COOKIE;
- } else {
- DBG_INIT("%s: No GART needed - no AGP controller found\n", __FUNCTION__);
+ ioc->pdir_base[PDIR_INDEX(iova_space_size/2)] = SBA_AGPGART_COOKIE;
}
-#endif /* 0 */
+}
+#endif /*SBA_AGP_SUPPORT*/
}
return single_open(f, &sba_proc_info, NULL);
}
-static struct file_operations sba_proc_fops = {
+static const struct file_operations sba_proc_fops = {
.owner = THIS_MODULE,
.open = sba_proc_open,
.read = seq_read,
return single_open(f, &sba_proc_bitmap_info, NULL);
}
-static struct file_operations sba_proc_bitmap_fops = {
+static const struct file_operations sba_proc_bitmap_fops = {
.owner = THIS_MODULE,
.open = sba_proc_bitmap_open,
.read = seq_read,
global_ioc_cnt *= 2;
}
- printk(KERN_INFO "%s found %s at 0x%lx\n",
- MODULE_NAME, version, dev->hpa.start);
+ printk(KERN_INFO "%s found %s at 0x%llx\n",
+ MODULE_NAME, version, (unsigned long long)dev->hpa.start);
sba_dev = kzalloc(sizeof(struct sba_device), GFP_KERNEL);
if (!sba_dev) {