Commit aebaa20d authored by Alex Williamson's avatar Alex Williamson Committed by David Mosberger

[PATCH] ia64: sba_iommu update

This patch does a couple things:
      * Allows iommu page size different than PAGE_SIZE.  This was
        largely done by Bjorn.  The motivation for this is that sx1000
        hardware officially only supports use of the iommu with a 4k
        page size.  Thus the default is now 4k for sx1000 and PAGE_SIZE
        for zx1.
      * Cleanup the runtime statistics gathering such that we can turn
        it on for some info w/o impacting runtime performance.
parent 58860a30
/* /*
** IA64 System Bus Adapter (SBA) I/O MMU manager ** IA64 System Bus Adapter (SBA) I/O MMU manager
** **
** (c) Copyright 2002-2003 Alex Williamson ** (c) Copyright 2002-2004 Alex Williamson
** (c) Copyright 2002-2003 Grant Grundler ** (c) Copyright 2002-2003 Grant Grundler
** (c) Copyright 2002-2003 Hewlett-Packard Company ** (c) Copyright 2002-2004 Hewlett-Packard Company
** **
** Portions (c) 2000 Grant Grundler (from parisc I/O MMU code) ** Portions (c) 2000 Grant Grundler (from parisc I/O MMU code)
** Portions (c) 1999 Dave S. Miller (from sparc64 I/O MMU code) ** Portions (c) 1999 Dave S. Miller (from sparc64 I/O MMU code)
...@@ -39,11 +39,18 @@ ...@@ -39,11 +39,18 @@
#include <asm/page.h> /* PAGE_OFFSET */ #include <asm/page.h> /* PAGE_OFFSET */
#include <asm/dma.h> #include <asm/dma.h>
#include <asm/system.h> /* wmb() */ #include <asm/system.h> /* wmb() */
#include <asm/bitops.h> /* hweight64() */
#include <asm/acpi-ext.h> #include <asm/acpi-ext.h>
#define PFX "IOC: " #define PFX "IOC: "
/*
** Enabling timing search of the pdir resource map. Output in /proc.
** Disabled by default to optimize performance.
*/
#undef PDIR_SEARCH_TIMING
/* /*
** This option allows cards capable of 64bit DMA to bypass the IOMMU. If ** This option allows cards capable of 64bit DMA to bypass the IOMMU. If
** not defined, all DMA will be 32bit and go through the TLB. ** not defined, all DMA will be 32bit and go through the TLB.
...@@ -54,11 +61,6 @@ ...@@ -54,11 +61,6 @@
*/ */
#define ALLOW_IOV_BYPASS #define ALLOW_IOV_BYPASS
#ifdef CONFIG_PROC_FS
/* turn it off for now; without per-CPU counters, it's too much of a scalability bottleneck: */
# define SBA_PROC_FS 0
#endif
/* /*
** If a device prefetches beyond the end of a valid pdir entry, it will cause ** If a device prefetches beyond the end of a valid pdir entry, it will cause
** a hard failure, ie. MCA. Version 3.0 and later of the zx1 LBA should ** a hard failure, ie. MCA. Version 3.0 and later of the zx1 LBA should
...@@ -160,21 +162,18 @@ ...@@ -160,21 +162,18 @@
#define ZX1_SBA_IOMMU_COOKIE 0x0000badbadc0ffeeUL #define ZX1_SBA_IOMMU_COOKIE 0x0000badbadc0ffeeUL
/* /*
** IOC supports 4/8/16/64KB page sizes (see TCNFG register) ** The zx1 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 ** Some IOCs (sx1000) can run at the above pages sizes, but are
** page since the Virtual Coherence Index has to be generated ** really only supported using the IOC at a 4k page size.
** and updated for each page.
** **
** IOVP_SIZE could only be greater than PAGE_SIZE if we are ** iovp_size could only be greater than PAGE_SIZE if we are
** confident the drivers really only touch the next physical ** confident the drivers really only touch the next physical
** page iff that driver instance owns it. ** page iff that driver instance owns it.
*/ */
#define IOVP_SIZE PAGE_SIZE static unsigned long iovp_size;
#define IOVP_SHIFT PAGE_SHIFT static unsigned long iovp_shift;
#define IOVP_MASK PAGE_MASK static unsigned long iovp_mask;
struct ioc { struct ioc {
void *ioc_hpa; /* I/O MMU base address */ void *ioc_hpa; /* I/O MMU base address */
...@@ -198,24 +197,10 @@ struct ioc { ...@@ -198,24 +197,10 @@ struct ioc {
} saved[DELAYED_RESOURCE_CNT]; } saved[DELAYED_RESOURCE_CNT];
#endif #endif
#if SBA_PROC_FS #ifdef PDIR_SEARCH_TIMING
#define SBA_SEARCH_SAMPLE 0x100 #define SBA_SEARCH_SAMPLE 0x100
unsigned long avg_search[SBA_SEARCH_SAMPLE]; unsigned long avg_search[SBA_SEARCH_SAMPLE];
unsigned long avg_idx; /* current index into avg_search */ 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;
#ifdef ALLOW_IOV_BYPASS
unsigned long msingle_bypass;
unsigned long usingle_bypass;
unsigned long msg_bypass;
#endif
#endif #endif
/* Stuff we don't need in performance path */ /* Stuff we don't need in performance path */
...@@ -252,7 +237,7 @@ static u64 prefetch_spill_page; ...@@ -252,7 +237,7 @@ static u64 prefetch_spill_page;
** rather than the HW. I/O MMU allocation alogorithms can be ** rather than the HW. I/O MMU allocation alogorithms can be
** faster with smaller size is (to some degree). ** faster with smaller size is (to some degree).
*/ */
#define DMA_CHUNK_SIZE (BITS_PER_LONG*PAGE_SIZE) #define DMA_CHUNK_SIZE (BITS_PER_LONG*iovp_size)
#define ROUNDUP(x,y) ((x + ((y)-1)) & ~((y)-1)) #define ROUNDUP(x,y) ((x + ((y)-1)) & ~((y)-1))
...@@ -416,17 +401,36 @@ sba_check_sg( struct ioc *ioc, struct scatterlist *startsg, int nents) ...@@ -416,17 +401,36 @@ sba_check_sg( struct ioc *ioc, struct scatterlist *startsg, int nents)
#define PAGES_PER_RANGE 1 /* could increase this to 4 or 8 if needed */ #define PAGES_PER_RANGE 1 /* could increase this to 4 or 8 if needed */
/* Convert from IOVP to IOVA and vice versa. */ /* Convert from IOVP to IOVA and vice versa. */
#define SBA_IOVA(ioc,iovp,offset,hint_reg) ((ioc->ibase) | (iovp) | (offset) | \ #define SBA_IOVA(ioc,iovp,offset,hint_reg) ((ioc->ibase) | (iovp) | (offset))
((hint_reg)<<(ioc->hint_shift_pdir))) #define SBA_IOVP(ioc,iova) ((iova) & ~(ioc->ibase))
#define SBA_IOVP(ioc,iova) (((iova) & ioc->hint_mask_pdir) & ~(ioc->ibase))
/* FIXME : review these macros to verify correctness and usage */ #define PDIR_ENTRY_SIZE sizeof(u64)
#define PDIR_INDEX(iovp) ((iovp)>>IOVP_SHIFT)
#define PDIR_INDEX(iovp) ((iovp)>>iovp_shift)
#define RESMAP_MASK(n) ~(~0UL << (n)) #define RESMAP_MASK(n) ~(~0UL << (n))
#define RESMAP_IDX_MASK (sizeof(unsigned long) - 1) #define RESMAP_IDX_MASK (sizeof(unsigned long) - 1)
/**
* For most cases the normal get_order is sufficient, however it limits us
* to PAGE_SIZE being the minimum mapping alignment and TC flush granularity.
* It only incurs about 1 clock cycle to use this one with the static variable
* and makes the code more intuitive.
*/
static SBA_INLINE int
get_iovp_order (unsigned long size)
{
long double d = size - 1;
long order;
order = ia64_getf_exp(d);
order = order - iovp_shift - 0xffff + 1;
if (order < 0)
order = 0;
return order;
}
/** /**
* sba_search_bitmap - find free space in IO PDIR resource bitmap * sba_search_bitmap - find free space in IO PDIR resource bitmap
* @ioc: IO MMU structure which owns the pdir we are interested in. * @ioc: IO MMU structure which owns the pdir we are interested in.
...@@ -465,7 +469,7 @@ sba_search_bitmap(struct ioc *ioc, unsigned long bits_wanted) ...@@ -465,7 +469,7 @@ sba_search_bitmap(struct ioc *ioc, unsigned long bits_wanted)
** We need the alignment to invalidate I/O TLB using ** We need the alignment to invalidate I/O TLB using
** SBA HW features in the unmap path. ** SBA HW features in the unmap path.
*/ */
unsigned long o = 1 << get_order(bits_wanted << PAGE_SHIFT); unsigned long o = 1 << get_iovp_order(bits_wanted << iovp_shift);
uint bitshiftcnt = ROUNDUP(ioc->res_bitshift, o); uint bitshiftcnt = ROUNDUP(ioc->res_bitshift, o);
unsigned long mask; unsigned long mask;
...@@ -521,16 +525,15 @@ sba_search_bitmap(struct ioc *ioc, unsigned long bits_wanted) ...@@ -521,16 +525,15 @@ sba_search_bitmap(struct ioc *ioc, unsigned long bits_wanted)
static int static int
sba_alloc_range(struct ioc *ioc, size_t size) sba_alloc_range(struct ioc *ioc, size_t size)
{ {
unsigned int pages_needed = size >> IOVP_SHIFT; unsigned int pages_needed = size >> iovp_shift;
#if SBA_PROC_FS #ifdef PDIR_SEARCH_TIMING
unsigned long itc_start = ia64_get_itc(); unsigned long itc_start = ia64_get_itc();
#endif #endif
unsigned long pide; unsigned long pide;
ASSERT(pages_needed); ASSERT(pages_needed);
ASSERT((pages_needed * IOVP_SIZE) <= DMA_CHUNK_SIZE);
ASSERT(pages_needed <= BITS_PER_LONG); ASSERT(pages_needed <= BITS_PER_LONG);
ASSERT(0 == (size & ~IOVP_MASK)); ASSERT(0 == (size & ~iovp_mask));
/* /*
** "seek and ye shall find"...praying never hurts either... ** "seek and ye shall find"...praying never hurts either...
...@@ -546,7 +549,7 @@ sba_alloc_range(struct ioc *ioc, size_t size) ...@@ -546,7 +549,7 @@ sba_alloc_range(struct ioc *ioc, size_t size)
#ifdef ASSERT_PDIR_SANITY #ifdef ASSERT_PDIR_SANITY
/* verify the first enable bit is clear */ /* verify the first enable bit is clear */
if(0x00 != ((u8 *) ioc->pdir_base)[pide*sizeof(u64) + 7]) { if(0x00 != ((u8 *) ioc->pdir_base)[pide*PDIR_ENTRY_SIZE + 7]) {
sba_dump_pdir_entry(ioc, "sba_search_bitmap() botched it?", pide); sba_dump_pdir_entry(ioc, "sba_search_bitmap() botched it?", pide);
} }
#endif #endif
...@@ -556,17 +559,9 @@ sba_alloc_range(struct ioc *ioc, size_t size) ...@@ -556,17 +559,9 @@ sba_alloc_range(struct ioc *ioc, size_t size)
(uint) ((unsigned long) ioc->res_hint - (unsigned long) ioc->res_map), (uint) ((unsigned long) ioc->res_hint - (unsigned long) ioc->res_map),
ioc->res_bitshift ); ioc->res_bitshift );
#if SBA_PROC_FS #ifdef PDIR_SEARCH_TIMING
{ ioc->avg_search[ioc->avg_idx++] = ia64_get_itc() - itc_start;
unsigned long itc_end = ia64_get_itc();
unsigned long tmp = itc_end - itc_start;
/* check for roll over */
itc_start = (itc_end < itc_start) ? -(tmp) : (tmp);
}
ioc->avg_search[ioc->avg_idx++] = itc_start;
ioc->avg_idx &= SBA_SEARCH_SAMPLE - 1; ioc->avg_idx &= SBA_SEARCH_SAMPLE - 1;
ioc->used_pages += pages_needed;
#endif #endif
return (pide); return (pide);
...@@ -589,7 +584,7 @@ sba_free_range(struct ioc *ioc, dma_addr_t iova, size_t size) ...@@ -589,7 +584,7 @@ sba_free_range(struct ioc *ioc, dma_addr_t iova, size_t size)
unsigned int ridx = pide >> 3; /* convert bit to byte address */ unsigned int ridx = pide >> 3; /* convert bit to byte address */
unsigned long *res_ptr = (unsigned long *) &((ioc)->res_map[ridx & ~RESMAP_IDX_MASK]); unsigned long *res_ptr = (unsigned long *) &((ioc)->res_map[ridx & ~RESMAP_IDX_MASK]);
int bits_not_wanted = size >> IOVP_SHIFT; int bits_not_wanted = size >> iovp_shift;
/* 3-bits "bit" address plus 2 (or 3) bits for "byte" == bit in word */ /* 3-bits "bit" address plus 2 (or 3) bits for "byte" == bit in word */
unsigned long m = RESMAP_MASK(bits_not_wanted) << (pide & (BITS_PER_LONG - 1)); unsigned long m = RESMAP_MASK(bits_not_wanted) << (pide & (BITS_PER_LONG - 1));
...@@ -598,13 +593,9 @@ sba_free_range(struct ioc *ioc, dma_addr_t iova, size_t size) ...@@ -598,13 +593,9 @@ sba_free_range(struct ioc *ioc, dma_addr_t iova, size_t size)
__FUNCTION__, (uint) iova, size, __FUNCTION__, (uint) iova, size,
bits_not_wanted, m, pide, res_ptr, *res_ptr); bits_not_wanted, m, pide, res_ptr, *res_ptr);
#if SBA_PROC_FS
ioc->used_pages -= bits_not_wanted;
#endif
ASSERT(m != 0); ASSERT(m != 0);
ASSERT(bits_not_wanted); ASSERT(bits_not_wanted);
ASSERT((bits_not_wanted * IOVP_SIZE) <= DMA_CHUNK_SIZE); ASSERT((bits_not_wanted * iovp_size) <= DMA_CHUNK_SIZE);
ASSERT(bits_not_wanted <= BITS_PER_LONG); ASSERT(bits_not_wanted <= BITS_PER_LONG);
ASSERT((*res_ptr & m) == m); /* verify same bits are set */ ASSERT((*res_ptr & m) == m); /* verify same bits are set */
*res_ptr &= ~m; *res_ptr &= ~m;
...@@ -702,7 +693,7 @@ sba_mark_invalid(struct ioc *ioc, dma_addr_t iova, size_t byte_cnt) ...@@ -702,7 +693,7 @@ sba_mark_invalid(struct ioc *ioc, dma_addr_t iova, size_t byte_cnt)
/* Must be non-zero and rounded up */ /* Must be non-zero and rounded up */
ASSERT(byte_cnt > 0); ASSERT(byte_cnt > 0);
ASSERT(0 == (byte_cnt & ~IOVP_MASK)); ASSERT(0 == (byte_cnt & ~iovp_mask));
#ifdef ASSERT_PDIR_SANITY #ifdef ASSERT_PDIR_SANITY
/* Assert first pdir entry is set */ /* Assert first pdir entry is set */
...@@ -711,11 +702,11 @@ sba_mark_invalid(struct ioc *ioc, dma_addr_t iova, size_t byte_cnt) ...@@ -711,11 +702,11 @@ sba_mark_invalid(struct ioc *ioc, dma_addr_t iova, size_t byte_cnt)
} }
#endif #endif
if (byte_cnt <= IOVP_SIZE) if (byte_cnt <= iovp_size)
{ {
ASSERT(off < ioc->pdir_size); ASSERT(off < ioc->pdir_size);
iovp |= IOVP_SHIFT; /* set "size" field for PCOM */ iovp |= iovp_shift; /* set "size" field for PCOM */
#ifndef FULL_VALID_PDIR #ifndef FULL_VALID_PDIR
/* /*
...@@ -734,7 +725,7 @@ sba_mark_invalid(struct ioc *ioc, dma_addr_t iova, size_t byte_cnt) ...@@ -734,7 +725,7 @@ sba_mark_invalid(struct ioc *ioc, dma_addr_t iova, size_t byte_cnt)
ioc->pdir_base[off] = (0x80000000000000FFULL | prefetch_spill_page); ioc->pdir_base[off] = (0x80000000000000FFULL | prefetch_spill_page);
#endif #endif
} else { } else {
u32 t = get_order(byte_cnt) + PAGE_SHIFT; u32 t = get_iovp_order(byte_cnt) + iovp_shift;
iovp |= t; iovp |= t;
ASSERT(t <= 31); /* 2GB! Max value of "size" field */ ASSERT(t <= 31); /* 2GB! Max value of "size" field */
...@@ -749,7 +740,7 @@ sba_mark_invalid(struct ioc *ioc, dma_addr_t iova, size_t byte_cnt) ...@@ -749,7 +740,7 @@ sba_mark_invalid(struct ioc *ioc, dma_addr_t iova, size_t byte_cnt)
ioc->pdir_base[off] = (0x80000000000000FFULL | prefetch_spill_page); ioc->pdir_base[off] = (0x80000000000000FFULL | prefetch_spill_page);
#endif #endif
off++; off++;
byte_cnt -= IOVP_SIZE; byte_cnt -= iovp_size;
} while (byte_cnt > 0); } while (byte_cnt > 0);
} }
...@@ -790,11 +781,6 @@ sba_map_single(struct device *dev, void *addr, size_t size, int dir) ...@@ -790,11 +781,6 @@ sba_map_single(struct device *dev, void *addr, size_t size, int dir)
** Device is bit capable of DMA'ing to the buffer... ** Device is bit capable of DMA'ing to the buffer...
** just return the PCI address of ptr ** just return the PCI address of ptr
*/ */
#if SBA_PROC_FS
spin_lock_irqsave(&ioc->res_lock, flags);
ioc->msingle_bypass++;
spin_unlock_irqrestore(&ioc->res_lock, flags);
#endif
DBG_BYPASS("sba_map_single() bypass mask/addr: 0x%lx/0x%lx\n", DBG_BYPASS("sba_map_single() bypass mask/addr: 0x%lx/0x%lx\n",
*dev->dma_mask, pci_addr); *dev->dma_mask, pci_addr);
return pci_addr; return pci_addr;
...@@ -805,10 +791,10 @@ sba_map_single(struct device *dev, void *addr, size_t size, int dir) ...@@ -805,10 +791,10 @@ sba_map_single(struct device *dev, void *addr, size_t size, int dir)
ASSERT(size <= DMA_CHUNK_SIZE); ASSERT(size <= DMA_CHUNK_SIZE);
/* save offset bits */ /* save offset bits */
offset = ((dma_addr_t) (long) addr) & ~IOVP_MASK; offset = ((dma_addr_t) (long) addr) & ~iovp_mask;
/* round up to nearest IOVP_SIZE */ /* round up to nearest iovp_size */
size = (size + offset + ~IOVP_MASK) & IOVP_MASK; size = (size + offset + ~iovp_mask) & iovp_mask;
spin_lock_irqsave(&ioc->res_lock, flags); spin_lock_irqsave(&ioc->res_lock, flags);
#ifdef ASSERT_PDIR_SANITY #ifdef ASSERT_PDIR_SANITY
...@@ -816,12 +802,8 @@ sba_map_single(struct device *dev, void *addr, size_t size, int dir) ...@@ -816,12 +802,8 @@ sba_map_single(struct device *dev, void *addr, size_t size, int dir)
panic("Sanity check failed"); panic("Sanity check failed");
#endif #endif
#if SBA_PROC_FS
ioc->msingle_calls++;
ioc->msingle_pages += size >> IOVP_SHIFT;
#endif
pide = sba_alloc_range(ioc, size); pide = sba_alloc_range(ioc, size);
iovp = (dma_addr_t) pide << IOVP_SHIFT; iovp = (dma_addr_t) pide << iovp_shift;
DBG_RUN("%s() 0x%p -> 0x%lx\n", DBG_RUN("%s() 0x%p -> 0x%lx\n",
__FUNCTION__, addr, (long) iovp | offset); __FUNCTION__, addr, (long) iovp | offset);
...@@ -834,8 +816,8 @@ sba_map_single(struct device *dev, void *addr, size_t size, int dir) ...@@ -834,8 +816,8 @@ sba_map_single(struct device *dev, void *addr, size_t size, int dir)
DBG_RUN(" pdir 0x%p %lx\n", pdir_start, *pdir_start); DBG_RUN(" pdir 0x%p %lx\n", pdir_start, *pdir_start);
addr += IOVP_SIZE; addr += iovp_size;
size -= IOVP_SIZE; size -= iovp_size;
pdir_start++; pdir_start++;
} }
/* force pdir update */ /* force pdir update */
...@@ -875,11 +857,6 @@ void sba_unmap_single(struct device *dev, dma_addr_t iova, size_t size, int dir) ...@@ -875,11 +857,6 @@ void sba_unmap_single(struct device *dev, dma_addr_t iova, size_t size, int dir)
/* /*
** Address does not fall w/in IOVA, must be bypassing ** Address does not fall w/in IOVA, must be bypassing
*/ */
#if SBA_PROC_FS
spin_lock_irqsave(&ioc->res_lock, flags);
ioc->usingle_bypass++;
spin_unlock_irqrestore(&ioc->res_lock, flags);
#endif
DBG_BYPASS("sba_unmap_single() bypass addr: 0x%lx\n", iova); DBG_BYPASS("sba_unmap_single() bypass addr: 0x%lx\n", iova);
#ifdef ENABLE_MARK_CLEAN #ifdef ENABLE_MARK_CLEAN
...@@ -890,20 +867,16 @@ void sba_unmap_single(struct device *dev, dma_addr_t iova, size_t size, int dir) ...@@ -890,20 +867,16 @@ void sba_unmap_single(struct device *dev, dma_addr_t iova, size_t size, int dir)
return; return;
} }
#endif #endif
offset = iova & ~IOVP_MASK; offset = iova & ~iovp_mask;
DBG_RUN("%s() iovp 0x%lx/%x\n", DBG_RUN("%s() iovp 0x%lx/%x\n",
__FUNCTION__, (long) iova, size); __FUNCTION__, (long) iova, size);
iova ^= offset; /* clear offset bits */ iova ^= offset; /* clear offset bits */
size += offset; size += offset;
size = ROUNDUP(size, IOVP_SIZE); size = ROUNDUP(size, iovp_size);
spin_lock_irqsave(&ioc->res_lock, flags); spin_lock_irqsave(&ioc->res_lock, flags);
#if SBA_PROC_FS
ioc->usingle_calls++;
ioc->usingle_pages += size >> IOVP_SHIFT;
#endif
#if DELAYED_RESOURCE_CNT > 0 #if DELAYED_RESOURCE_CNT > 0
d = &(ioc->saved[ioc->saved_cnt]); d = &(ioc->saved[ioc->saved_cnt]);
...@@ -930,7 +903,7 @@ void sba_unmap_single(struct device *dev, dma_addr_t iova, size_t size, int dir) ...@@ -930,7 +903,7 @@ void sba_unmap_single(struct device *dev, dma_addr_t iova, size_t size, int dir)
int off = PDIR_INDEX(iovp); int off = PDIR_INDEX(iovp);
void *addr; void *addr;
if (size <= IOVP_SIZE) { if (size <= iovp_size) {
addr = phys_to_virt(ioc->pdir_base[off] & addr = phys_to_virt(ioc->pdir_base[off] &
~0xE000000000000FFFULL); ~0xE000000000000FFFULL);
mark_clean(addr, size); mark_clean(addr, size);
...@@ -940,9 +913,9 @@ void sba_unmap_single(struct device *dev, dma_addr_t iova, size_t size, int dir) ...@@ -940,9 +913,9 @@ void sba_unmap_single(struct device *dev, dma_addr_t iova, size_t size, int dir)
do { do {
addr = phys_to_virt(ioc->pdir_base[off] & addr = phys_to_virt(ioc->pdir_base[off] &
~0xE000000000000FFFULL); ~0xE000000000000FFFULL);
mark_clean(addr, min(byte_cnt, IOVP_SIZE)); mark_clean(addr, min(byte_cnt, iovp_size));
off++; off++;
byte_cnt -= IOVP_SIZE; byte_cnt -= iovp_size;
} while (byte_cnt > 0); } while (byte_cnt > 0);
} }
...@@ -1061,11 +1034,11 @@ sba_fill_pdir( ...@@ -1061,11 +1034,11 @@ sba_fill_pdir(
*/ */
if (startsg->dma_address & PIDE_FLAG) { if (startsg->dma_address & PIDE_FLAG) {
u32 pide = startsg->dma_address & ~PIDE_FLAG; u32 pide = startsg->dma_address & ~PIDE_FLAG;
dma_offset = (unsigned long) pide & ~IOVP_MASK; dma_offset = (unsigned long) pide & ~iovp_mask;
startsg->dma_address = 0; startsg->dma_address = 0;
dma_sg++; dma_sg++;
dma_sg->dma_address = pide | ioc->ibase; dma_sg->dma_address = pide | ioc->ibase;
pdirp = &(ioc->pdir_base[pide >> IOVP_SHIFT]); pdirp = &(ioc->pdir_base[pide >> iovp_shift]);
n_mappings++; n_mappings++;
} }
...@@ -1082,14 +1055,11 @@ sba_fill_pdir( ...@@ -1082,14 +1055,11 @@ sba_fill_pdir(
dma_sg->dma_length += cnt; dma_sg->dma_length += cnt;
cnt += dma_offset; cnt += dma_offset;
dma_offset=0; /* only want offset on first chunk */ dma_offset=0; /* only want offset on first chunk */
cnt = ROUNDUP(cnt, IOVP_SIZE); cnt = ROUNDUP(cnt, iovp_size);
#if SBA_PROC_FS
ioc->msg_pages += cnt >> IOVP_SHIFT;
#endif
do { do {
sba_io_pdir_entry(pdirp, vaddr); sba_io_pdir_entry(pdirp, vaddr);
vaddr += IOVP_SIZE; vaddr += iovp_size;
cnt -= IOVP_SIZE; cnt -= iovp_size;
pdirp++; pdirp++;
} while (cnt > 0); } while (cnt > 0);
} }
...@@ -1107,12 +1077,12 @@ sba_fill_pdir( ...@@ -1107,12 +1077,12 @@ sba_fill_pdir(
/* /*
** Two address ranges are DMA contiguous *iff* "end of prev" and ** Two address ranges are DMA contiguous *iff* "end of prev" and
** "start of next" are both on a page boundry. ** "start of next" are both on an IOV page boundary.
** **
** (shift left is a quick trick to mask off upper bits) ** (shift left is a quick trick to mask off upper bits)
*/ */
#define DMA_CONTIG(__X, __Y) \ #define DMA_CONTIG(__X, __Y) \
(((((unsigned long) __X) | ((unsigned long) __Y)) << (BITS_PER_LONG - PAGE_SHIFT)) == 0UL) (((((unsigned long) __X) | ((unsigned long) __Y)) << (BITS_PER_LONG - iovp_shift)) == 0UL)
/** /**
...@@ -1150,7 +1120,7 @@ sba_coalesce_chunks( struct ioc *ioc, ...@@ -1150,7 +1120,7 @@ sba_coalesce_chunks( struct ioc *ioc,
dma_sg = vcontig_sg = startsg; dma_sg = vcontig_sg = startsg;
dma_len = vcontig_len = vcontig_end = startsg->length; dma_len = vcontig_len = vcontig_end = startsg->length;
vcontig_end += vaddr; vcontig_end += vaddr;
dma_offset = vaddr & ~IOVP_MASK; dma_offset = vaddr & ~iovp_mask;
/* PARANOID: clear entries */ /* PARANOID: clear entries */
startsg->dma_address = startsg->dma_length = 0; startsg->dma_address = startsg->dma_length = 0;
...@@ -1175,7 +1145,7 @@ sba_coalesce_chunks( struct ioc *ioc, ...@@ -1175,7 +1145,7 @@ sba_coalesce_chunks( struct ioc *ioc,
** exceed DMA_CHUNK_SIZE if we coalesce the ** exceed DMA_CHUNK_SIZE if we coalesce the
** next entry. ** next entry.
*/ */
if (((dma_len + dma_offset + startsg->length + ~IOVP_MASK) & IOVP_MASK) if (((dma_len + dma_offset + startsg->length + ~iovp_mask) & iovp_mask)
> DMA_CHUNK_SIZE) > DMA_CHUNK_SIZE)
break; break;
...@@ -1194,7 +1164,7 @@ sba_coalesce_chunks( struct ioc *ioc, ...@@ -1194,7 +1164,7 @@ sba_coalesce_chunks( struct ioc *ioc,
} }
#ifdef DEBUG_LARGE_SG_ENTRIES #ifdef DEBUG_LARGE_SG_ENTRIES
dump_run_sg = (vcontig_len > IOVP_SIZE); dump_run_sg = (vcontig_len > iovp_size);
#endif #endif
/* /*
...@@ -1233,10 +1203,10 @@ sba_coalesce_chunks( struct ioc *ioc, ...@@ -1233,10 +1203,10 @@ sba_coalesce_chunks( struct ioc *ioc,
** Allocate space for DMA stream. ** Allocate space for DMA stream.
*/ */
vcontig_sg->dma_length = vcontig_len; vcontig_sg->dma_length = vcontig_len;
dma_len = (dma_len + dma_offset + ~IOVP_MASK) & IOVP_MASK; dma_len = (dma_len + dma_offset + ~iovp_mask) & iovp_mask;
ASSERT(dma_len <= DMA_CHUNK_SIZE); ASSERT(dma_len <= DMA_CHUNK_SIZE);
dma_sg->dma_address = (dma_addr_t) (PIDE_FLAG dma_sg->dma_address = (dma_addr_t) (PIDE_FLAG
| (sba_alloc_range(ioc, dma_len) << IOVP_SHIFT) | (sba_alloc_range(ioc, dma_len) << iovp_shift)
| dma_offset); | dma_offset);
n_mappings++; n_mappings++;
} }
...@@ -1273,11 +1243,6 @@ int sba_map_sg(struct device *dev, struct scatterlist *sglist, int nents, int di ...@@ -1273,11 +1243,6 @@ int sba_map_sg(struct device *dev, struct scatterlist *sglist, int nents, int di
sg->dma_length = sg->length; sg->dma_length = sg->length;
sg->dma_address = virt_to_phys(sba_sg_address(sg)); sg->dma_address = virt_to_phys(sba_sg_address(sg));
} }
#if SBA_PROC_FS
spin_lock_irqsave(&ioc->res_lock, flags);
ioc->msg_bypass++;
spin_unlock_irqrestore(&ioc->res_lock, flags);
#endif
return filled; return filled;
} }
#endif #endif
...@@ -1286,12 +1251,6 @@ int sba_map_sg(struct device *dev, struct scatterlist *sglist, int nents, int di ...@@ -1286,12 +1251,6 @@ int sba_map_sg(struct device *dev, struct scatterlist *sglist, int nents, int di
sglist->dma_length = sglist->length; sglist->dma_length = sglist->length;
sglist->dma_address = sba_map_single(dev, sba_sg_address(sglist), sglist->length, sglist->dma_address = sba_map_single(dev, sba_sg_address(sglist), sglist->length,
dir); dir);
#if SBA_PROC_FS
/*
** Should probably do some stats counting, but trying to
** be precise quickly starts wasting CPU time.
*/
#endif
return 1; return 1;
} }
...@@ -1305,10 +1264,6 @@ int sba_map_sg(struct device *dev, struct scatterlist *sglist, int nents, int di ...@@ -1305,10 +1264,6 @@ int sba_map_sg(struct device *dev, struct scatterlist *sglist, int nents, int di
} }
#endif #endif
#if SBA_PROC_FS
ioc->msg_calls++;
#endif
/* /*
** First coalesce the chunks and allocate I/O pdir space ** First coalesce the chunks and allocate I/O pdir space
** **
...@@ -1368,10 +1323,6 @@ void sba_unmap_sg (struct device *dev, struct scatterlist *sglist, int nents, in ...@@ -1368,10 +1323,6 @@ void sba_unmap_sg (struct device *dev, struct scatterlist *sglist, int nents, in
ioc = GET_IOC(dev); ioc = GET_IOC(dev);
ASSERT(ioc); ASSERT(ioc);
#if SBA_PROC_FS
ioc->usg_calls++;
#endif
#ifdef ASSERT_PDIR_SANITY #ifdef ASSERT_PDIR_SANITY
spin_lock_irqsave(&ioc->res_lock, flags); spin_lock_irqsave(&ioc->res_lock, flags);
sba_check_pdir(ioc,"Check before sba_unmap_sg()"); sba_check_pdir(ioc,"Check before sba_unmap_sg()");
...@@ -1381,16 +1332,6 @@ void sba_unmap_sg (struct device *dev, struct scatterlist *sglist, int nents, in ...@@ -1381,16 +1332,6 @@ void sba_unmap_sg (struct device *dev, struct scatterlist *sglist, int nents, in
while (nents && sglist->dma_length) { while (nents && sglist->dma_length) {
sba_unmap_single(dev, sglist->dma_address, sglist->dma_length, dir); sba_unmap_single(dev, sglist->dma_address, sglist->dma_length, dir);
#if SBA_PROC_FS
/*
** This leaves inconsistent data in the stats, but we can't
** tell which sg lists were mapped by map_single and which
** were coalesced to a single entry. The stats are fun,
** but speed is more important.
*/
ioc->usg_pages += ((sglist->dma_address & ~IOVP_MASK) + sglist->dma_length
+ IOVP_SIZE - 1) >> PAGE_SHIFT;
#endif
sglist++; sglist++;
nents--; nents--;
} }
...@@ -1414,8 +1355,7 @@ void sba_unmap_sg (struct device *dev, struct scatterlist *sglist, int nents, in ...@@ -1414,8 +1355,7 @@ void sba_unmap_sg (struct device *dev, struct scatterlist *sglist, int nents, in
static void __init static void __init
ioc_iova_init(struct ioc *ioc) ioc_iova_init(struct ioc *ioc)
{ {
u32 iova_space_mask; int tcnfg;
int iov_order, tcnfg;
int agp_found = 0; int agp_found = 0;
struct pci_dev *device = NULL; struct pci_dev *device = NULL;
#ifdef FULL_VALID_PDIR #ifdef FULL_VALID_PDIR
...@@ -1428,23 +1368,27 @@ ioc_iova_init(struct ioc *ioc) ...@@ -1428,23 +1368,27 @@ ioc_iova_init(struct ioc *ioc)
** IBASE and IMASK registers. ** IBASE and IMASK registers.
*/ */
ioc->ibase = READ_REG(ioc->ioc_hpa + IOC_IBASE) & ~0x1UL; ioc->ibase = READ_REG(ioc->ioc_hpa + IOC_IBASE) & ~0x1UL;
ioc->iov_size = ~(READ_REG(ioc->ioc_hpa + IOC_IMASK) & 0xFFFFFFFFUL) + 1; ioc->imask = READ_REG(ioc->ioc_hpa + IOC_IMASK) | 0xFFFFFFFF00000000UL;
/* ioc->iov_size = ~ioc->imask + 1;
** iov_order is always based on a 1GB IOVA space since we want to
** turn on the other half for AGP GART.
*/
iov_order = get_order(ioc->iov_size >> (IOVP_SHIFT - PAGE_SHIFT));
ioc->pdir_size = (ioc->iov_size / IOVP_SIZE) * sizeof(u64);
DBG_INIT("%s() hpa %p IOV %dMB (%d bits) PDIR size 0x%x\n", DBG_INIT("%s() hpa %p IOV base 0x%lx mask 0x%lx (%dMB)\n",
__FUNCTION__, ioc->ioc_hpa, ioc->iov_size >> 20, __FUNCTION__, ioc->ioc_hpa, ioc->ibase, ioc->imask,
iov_order + PAGE_SHIFT, ioc->pdir_size); ioc->iov_size >> 20);
/* FIXME : DMA HINTs not used */ switch (iovp_size) {
ioc->hint_shift_pdir = iov_order + PAGE_SHIFT; case 4*1024: tcnfg = 0; break;
ioc->hint_mask_pdir = ~(0x3 << (iov_order + PAGE_SHIFT)); case 8*1024: tcnfg = 1; break;
case 16*1024: tcnfg = 2; break;
case 64*1024: tcnfg = 3; break;
default:
panic(PFX "Unsupported IOTLB page size %ldK",
iovp_size >> 10);
break;
}
WRITE_REG(tcnfg, ioc->ioc_hpa + IOC_TCNFG);
ioc->pdir_size = (ioc->iov_size / iovp_size) * PDIR_ENTRY_SIZE;
ioc->pdir_base = (void *) __get_free_pages(GFP_KERNEL, ioc->pdir_base = (void *) __get_free_pages(GFP_KERNEL,
get_order(ioc->pdir_size)); get_order(ioc->pdir_size));
if (!ioc->pdir_base) if (!ioc->pdir_base)
...@@ -1452,61 +1396,12 @@ ioc_iova_init(struct ioc *ioc) ...@@ -1452,61 +1396,12 @@ ioc_iova_init(struct ioc *ioc)
memset(ioc->pdir_base, 0, ioc->pdir_size); memset(ioc->pdir_base, 0, ioc->pdir_size);
DBG_INIT("%s() pdir %p size %x hint_shift_pdir %x hint_mask_pdir %lx\n", DBG_INIT("%s() IOV page size %ldK pdir %p size %x\n", __FUNCTION__,
__FUNCTION__, ioc->pdir_base, ioc->pdir_size, iovp_size >> 10, ioc->pdir_base, ioc->pdir_size);
ioc->hint_shift_pdir, ioc->hint_mask_pdir);
ASSERT((((unsigned long) ioc->pdir_base) & PAGE_MASK) == (unsigned long) ioc->pdir_base); ASSERT(ALIGN((unsigned long) ioc->pdir_base, 4*1024) == (unsigned long) ioc->pdir_base);
WRITE_REG(virt_to_phys(ioc->pdir_base), ioc->ioc_hpa + IOC_PDIR_BASE); WRITE_REG(virt_to_phys(ioc->pdir_base), ioc->ioc_hpa + IOC_PDIR_BASE);
DBG_INIT(" base %p\n", ioc->pdir_base);
/* build IMASK for IOC and Elroy */
iova_space_mask = 0xffffffff;
iova_space_mask <<= (iov_order + PAGE_SHIFT);
ioc->imask = iova_space_mask;
DBG_INIT("%s() IOV base 0x%lx mask 0x%0lx\n",
__FUNCTION__, ioc->ibase, ioc->imask);
/*
** FIXME: Hint registers are programmed with default hint
** values during boot, so hints should be sane even if we
** can't reprogram them the way drivers want.
*/
WRITE_REG(ioc->imask, ioc->ioc_hpa + IOC_IMASK);
/*
** Setting the upper bits makes checking for bypass addresses
** a little faster later on.
*/
ioc->imask |= 0xFFFFFFFF00000000UL;
/* Set I/O PDIR Page size to system page size */
switch (PAGE_SHIFT) {
case 12: tcnfg = 0; break; /* 4K */
case 13: tcnfg = 1; break; /* 8K */
case 14: tcnfg = 2; break; /* 16K */
case 16: tcnfg = 3; break; /* 64K */
default:
panic(PFX "Unsupported system page size %d",
1 << PAGE_SHIFT);
break;
}
WRITE_REG(tcnfg, ioc->ioc_hpa + IOC_TCNFG);
/*
** Program the IOC's ibase and enable IOVA translation
** Bit zero == enable bit.
*/
WRITE_REG(ioc->ibase | 1, ioc->ioc_hpa + IOC_IBASE);
/*
** Clear I/O TLB of any possible entries.
** (Yes. This is a bit paranoid...but so what)
*/
WRITE_REG(ioc->ibase | (iov_order+PAGE_SHIFT), ioc->ioc_hpa + IOC_PCOM);
/* /*
** If an AGP device is present, only use half of the IOV space ** If an AGP device is present, only use half of the IOV space
** for PCI DMA. Unfortunately we can't know ahead of time ** for PCI DMA. Unfortunately we can't know ahead of time
...@@ -1534,12 +1429,12 @@ ioc_iova_init(struct ioc *ioc) ...@@ -1534,12 +1429,12 @@ ioc_iova_init(struct ioc *ioc)
int poison_size = 16; int poison_size = 16;
void *poison_addr, *addr; void *poison_addr, *addr;
addr = (void *)__get_free_pages(GFP_KERNEL, get_order(IOVP_SIZE)); addr = (void *)__get_free_pages(GFP_KERNEL, get_order(iovp_size));
if (!addr) if (!addr)
panic(PFX "Couldn't allocate PDIR spill page\n"); panic(PFX "Couldn't allocate PDIR spill page\n");
poison_addr = addr; poison_addr = addr;
for ( ; (u64) poison_addr < addr + IOVP_SIZE; poison_addr += poison_size) for ( ; (u64) poison_addr < addr + iovp_size; poison_addr += poison_size)
memcpy(poison_addr, spill_poison, poison_size); memcpy(poison_addr, spill_poison, poison_size);
prefetch_spill_page = virt_to_phys(addr); prefetch_spill_page = virt_to_phys(addr);
...@@ -1549,10 +1444,17 @@ ioc_iova_init(struct ioc *ioc) ...@@ -1549,10 +1444,17 @@ ioc_iova_init(struct ioc *ioc)
/* /*
** Set all the PDIR entries valid w/ the spill page as the target ** Set all the PDIR entries valid w/ the spill page as the target
*/ */
for (index = 0 ; index < (ioc->pdir_size / sizeof(u64)) ; index++) for (index = 0 ; index < (ioc->pdir_size / PDIR_ENTRY_SIZE) ; index++)
((u64 *)ioc->pdir_base)[index] = (0x80000000000000FF | prefetch_spill_page); ((u64 *)ioc->pdir_base)[index] = (0x80000000000000FF | prefetch_spill_page);
#endif #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 static void __init
...@@ -1561,7 +1463,7 @@ ioc_resource_init(struct ioc *ioc) ...@@ -1561,7 +1463,7 @@ ioc_resource_init(struct ioc *ioc)
spin_lock_init(&ioc->res_lock); spin_lock_init(&ioc->res_lock);
/* resource map size dictated by pdir_size */ /* resource map size dictated by pdir_size */
ioc->res_size = ioc->pdir_size / sizeof(u64); /* entries */ ioc->res_size = ioc->pdir_size / PDIR_ENTRY_SIZE; /* entries */
ioc->res_size >>= 3; /* convert bit count to byte count */ ioc->res_size >>= 3; /* convert bit count to byte count */
DBG_INIT("%s() res_size 0x%x\n", __FUNCTION__, ioc->res_size); DBG_INIT("%s() res_size 0x%x\n", __FUNCTION__, ioc->res_size);
...@@ -1582,7 +1484,7 @@ ioc_resource_init(struct ioc *ioc) ...@@ -1582,7 +1484,7 @@ ioc_resource_init(struct ioc *ioc)
#ifdef FULL_VALID_PDIR #ifdef FULL_VALID_PDIR
/* Mark the last resource used so we don't prefetch beyond IOVA space */ /* 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->res_map[ioc->res_size - 1] |= 0x80UL; /* res_map is chars */
ioc->pdir_base[(ioc->pdir_size / sizeof(u64)) - 1] = (0x80000000000000FF ioc->pdir_base[(ioc->pdir_size / PDIR_ENTRY_SIZE) - 1] = (0x80000000000000FF
| prefetch_spill_page); | prefetch_spill_page);
#endif #endif
...@@ -1627,6 +1529,23 @@ ioc_zx1_init(struct ioc *ioc) ...@@ -1627,6 +1529,23 @@ ioc_zx1_init(struct ioc *ioc)
panic(PFX "IOC 2.0 or later required for IOMMU support\n"); panic(PFX "IOC 2.0 or later required for IOMMU support\n");
ioc->dma_mask = 0xFFFFFFFFFFUL; ioc->dma_mask = 0xFFFFFFFFFFUL;
if (!iovp_shift) {
/* 64k is max iommu page size */
iovp_shift = min(PAGE_SHIFT, 16);
iovp_size = (1 << iovp_shift);
iovp_mask = ~(iovp_size - 1);
}
}
static void __init
ioc_sx1000_init(struct ioc *ioc)
{
if (!iovp_shift) {
iovp_shift = 12; /* 4K for now */
iovp_size = (1 << iovp_shift);
iovp_mask = ~(iovp_size - 1);
}
} }
typedef void (initfunc)(struct ioc *); typedef void (initfunc)(struct ioc *);
...@@ -1639,8 +1558,8 @@ struct ioc_iommu { ...@@ -1639,8 +1558,8 @@ struct ioc_iommu {
static struct ioc_iommu ioc_iommu_info[] __initdata = { static struct ioc_iommu ioc_iommu_info[] __initdata = {
{ ZX1_IOC_ID, "zx1", ioc_zx1_init }, { ZX1_IOC_ID, "zx1", ioc_zx1_init },
{ REO_IOC_ID, "REO" }, { REO_IOC_ID, "REO", ioc_sx1000_init },
{ SX1000_IOC_ID, "sx1000" }, { SX1000_IOC_ID, "sx1000", ioc_sx1000_init },
}; };
static struct ioc * __init static struct ioc * __init
...@@ -1665,6 +1584,11 @@ ioc_init(u64 hpa, void *handle) ...@@ -1665,6 +1584,11 @@ ioc_init(u64 hpa, void *handle)
ioc->rev = READ_REG(ioc->ioc_hpa + IOC_FCLASS) & 0xFFUL; ioc->rev = READ_REG(ioc->ioc_hpa + IOC_FCLASS) & 0xFFUL;
ioc->dma_mask = 0xFFFFFFFFFFFFFFFFUL; /* conservative */ ioc->dma_mask = 0xFFFFFFFFFFFFFFFFUL; /* conservative */
if (iovp_shift) {
iovp_size = (1 << iovp_shift);
iovp_mask = ~(iovp_size - 1);
}
for (info = ioc_iommu_info; info < ioc_iommu_info + ARRAY_SIZE(ioc_iommu_info); info++) { for (info = ioc_iommu_info; info < ioc_iommu_info + ARRAY_SIZE(ioc_iommu_info); info++) {
if (ioc->func_id == info->func_id) { if (ioc->func_id == info->func_id) {
ioc->name = info->name; ioc->name = info->name;
...@@ -1672,6 +1596,8 @@ ioc_init(u64 hpa, void *handle) ...@@ -1672,6 +1596,8 @@ ioc_init(u64 hpa, void *handle)
(info->init)(ioc); (info->init)(ioc);
} }
} }
DBG_INIT("%s: PAGE_SIZE %ldK, iovp_size %ldK\n", __FUNCTION__,
PAGE_SIZE >> 10, iovp_size >> 10);
if (!ioc->name) { if (!ioc->name) {
ioc->name = kmalloc(24, GFP_KERNEL); ioc->name = kmalloc(24, GFP_KERNEL);
...@@ -1686,8 +1612,8 @@ ioc_init(u64 hpa, void *handle) ...@@ -1686,8 +1612,8 @@ ioc_init(u64 hpa, void *handle)
ioc_resource_init(ioc); ioc_resource_init(ioc);
ioc_sac_init(ioc); ioc_sac_init(ioc);
if ((long) ~IOVP_MASK > (long) ia64_max_iommu_merge_mask) if ((long) ~iovp_mask > (long) ia64_max_iommu_merge_mask)
ia64_max_iommu_merge_mask = ~IOVP_MASK; ia64_max_iommu_merge_mask = ~iovp_mask;
MAX_DMA_ADDRESS = ~0UL; MAX_DMA_ADDRESS = ~0UL;
printk(KERN_INFO PFX printk(KERN_INFO PFX
...@@ -1709,7 +1635,7 @@ ioc_init(u64 hpa, void *handle) ...@@ -1709,7 +1635,7 @@ ioc_init(u64 hpa, void *handle)
** **
**************************************************************************/ **************************************************************************/
#if SBA_PROC_FS #ifdef CONFIG_PROC_FS
static void * static void *
ioc_start(struct seq_file *s, loff_t *pos) ioc_start(struct seq_file *s, loff_t *pos)
{ {
...@@ -1741,22 +1667,23 @@ static int ...@@ -1741,22 +1667,23 @@ static int
ioc_show(struct seq_file *s, void *v) ioc_show(struct seq_file *s, void *v)
{ {
struct ioc *ioc = v; struct ioc *ioc = v;
int total_pages = (int) (ioc->res_size << 3); /* 8 bits per byte */ unsigned long *res_ptr = (unsigned long *)ioc->res_map;
unsigned long i = 0, avg = 0, min, max; int i, used = 0;
seq_printf(s, "Hewlett Packard %s IOC rev %d.%d\n", seq_printf(s, "Hewlett Packard %s IOC rev %d.%d\n",
ioc->name, ((ioc->rev >> 4) & 0xF), (ioc->rev & 0xF)); ioc->name, ((ioc->rev >> 4) & 0xF), (ioc->rev & 0xF));
seq_printf(s, "IO PDIR size : %d bytes (%d entries)\n", seq_printf(s, "IOVA size : %d MB\n", ioc->iov_size/(1024*1024));
(int) ((ioc->res_size << 3) * sizeof(u64)), /* 8 bits/byte */ seq_printf(s, "IOVA page size : %ld kb\n", iovp_size/1024);
total_pages);
seq_printf(s, "IO PDIR entries : %ld free %ld used (%d%%)\n", for (i = 0; i < (ioc->res_size / sizeof(unsigned long)); ++i, ++res_ptr)
total_pages - ioc->used_pages, ioc->used_pages, used += hweight64(*res_ptr);
(int) (ioc->used_pages * 100 / total_pages));
seq_printf(s, "Resource bitmap : %d bytes (%d pages)\n", seq_printf(s, "PDIR size : %d entries\n", ioc->res_size << 3);
ioc->res_size, ioc->res_size << 3); /* 8 bits per byte */ 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]; min = max = ioc->avg_search[0];
for (i = 0; i < SBA_SEARCH_SAMPLE; i++) { for (i = 0; i < SBA_SEARCH_SAMPLE; i++) {
avg += ioc->avg_search[i]; avg += ioc->avg_search[i];
...@@ -1764,32 +1691,13 @@ ioc_show(struct seq_file *s, void *v) ...@@ -1764,32 +1691,13 @@ ioc_show(struct seq_file *s, void *v)
if (ioc->avg_search[i] < min) min = ioc->avg_search[i]; if (ioc->avg_search[i] < min) min = ioc->avg_search[i];
} }
avg /= SBA_SEARCH_SAMPLE; avg /= SBA_SEARCH_SAMPLE;
seq_printf(s, " Bitmap search : %ld/%ld/%ld (min/avg/max CPU Cycles)\n", min, avg, max); seq_printf(s, "Bitmap search : %ld/%ld/%ld (min/avg/max CPU Cycles)\n",
min, avg, max);
seq_printf(s, "pci_map_single(): %12ld calls %12ld pages (avg %d/1000)\n", }
ioc->msingle_calls, ioc->msingle_pages,
(int) ((ioc->msingle_pages * 1000)/ioc->msingle_calls));
#ifdef ALLOW_IOV_BYPASS
seq_printf(s, "pci_map_single(): %12ld bypasses\n", ioc->msingle_bypass);
#endif
seq_printf(s, "pci_unmap_single: %12ld calls %12ld pages (avg %d/1000)\n",
ioc->usingle_calls, ioc->usingle_pages,
(int) ((ioc->usingle_pages * 1000)/ioc->usingle_calls));
#ifdef ALLOW_IOV_BYPASS
seq_printf(s, "pci_unmap_single: %12ld bypasses\n", ioc->usingle_bypass);
#endif #endif
#ifndef ALLOW_IOV_BYPASS
seq_printf(s, "pci_map_sg() : %12ld calls %12ld pages (avg %d/1000)\n", seq_printf(s, "IOVA bypass disabled\n");
ioc->msg_calls, ioc->msg_pages,
(int) ((ioc->msg_pages * 1000)/ioc->msg_calls));
#ifdef ALLOW_IOV_BYPASS
seq_printf(s, "pci_map_sg() : %12ld bypasses\n", ioc->msg_bypass);
#endif #endif
seq_printf(s, "pci_unmap_sg() : %12ld calls %12ld pages (avg %d/1000)\n",
ioc->usg_calls, ioc->usg_pages, (int) ((ioc->usg_pages * 1000)/ioc->usg_calls));
return 0; return 0;
} }
...@@ -1813,39 +1721,6 @@ static struct file_operations ioc_fops = { ...@@ -1813,39 +1721,6 @@ static struct file_operations ioc_fops = {
.release = seq_release .release = seq_release
}; };
static int
ioc_map_show(struct seq_file *s, void *v)
{
struct ioc *ioc = v;
unsigned int i, *res_ptr = (unsigned int *)ioc->res_map;
for (i = 0; i < ioc->res_size / sizeof(unsigned int); ++i, ++res_ptr)
seq_printf(s, "%s%08x", (i & 7) ? " " : "\n ", *res_ptr);
seq_printf(s, "\n");
return 0;
}
static struct seq_operations ioc_map_ops = {
.start = ioc_start,
.next = ioc_next,
.stop = ioc_stop,
.show = ioc_map_show
};
static int
ioc_map_open(struct inode *inode, struct file *file)
{
return seq_open(file, &ioc_map_ops);
}
static struct file_operations ioc_map_fops = {
.open = ioc_map_open,
.read = seq_read,
.llseek = seq_lseek,
.release = seq_release
};
static void __init static void __init
ioc_proc_init(void) ioc_proc_init(void)
{ {
...@@ -1858,10 +1733,6 @@ ioc_proc_init(void) ...@@ -1858,10 +1733,6 @@ ioc_proc_init(void)
entry = create_proc_entry(ioc_list->name, 0, dir); entry = create_proc_entry(ioc_list->name, 0, dir);
if (entry) if (entry)
entry->proc_fops = &ioc_fops; entry->proc_fops = &ioc_fops;
entry = create_proc_entry("bitmap", 0, dir);
if (entry)
entry->proc_fops = &ioc_map_fops;
} }
#endif #endif
...@@ -1958,7 +1829,7 @@ sba_init(void) ...@@ -1958,7 +1829,7 @@ sba_init(void)
} }
#endif #endif
#if SBA_PROC_FS #ifdef CONFIG_PROC_FS
ioc_proc_init(); ioc_proc_init();
#endif #endif
return 0; return 0;
...@@ -1982,6 +1853,29 @@ sba_dma_supported (struct device *dev, u64 mask) ...@@ -1982,6 +1853,29 @@ sba_dma_supported (struct device *dev, u64 mask)
__setup("nosbagart", nosbagart); __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_map_single); EXPORT_SYMBOL(sba_map_single);
EXPORT_SYMBOL(sba_unmap_single); EXPORT_SYMBOL(sba_unmap_single);
EXPORT_SYMBOL(sba_map_sg); EXPORT_SYMBOL(sba_map_sg);
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment