[PATCH] x86_64: Calgary IOMMU - Calgary specific bits
authorJon Mason <jdmason@us.ibm.com>
Mon, 26 Jun 2006 11:58:14 +0000 (13:58 +0200)
committerLinus Torvalds <torvalds@g5.osdl.org>
Mon, 26 Jun 2006 17:48:19 +0000 (10:48 -0700)
This patch hooks Calgary into the build, the x86-64 IOMMU
initialization paths, and introduces the Calgary specific bits.  The
implementation draws inspiration from both PPC (which has support for
the same chip but requires firmware support which we don't have on
x86-64) and gart. Calgary is different from gart in that it support a
translation table per PHB, as opposed to the single gart aperture.

Changes from previous version:
 * Addition of boot-time disablement for bus-level translation/isolation
   (e.g, enable userspace DMA for things like X)
 * Usage of newer IOMMU abstraction functions

Signed-off-by: Muli Ben-Yehuda <muli@il.ibm.com>
Signed-off-by: Jon Mason <jdmason@us.ibm.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Documentation/x86_64/boot-options.txt
arch/x86_64/Kconfig
arch/x86_64/kernel/Makefile
arch/x86_64/kernel/pci-calgary.c [new file with mode: 0644]
arch/x86_64/kernel/pci-dma.c
arch/x86_64/kernel/tce.c [new file with mode: 0644]
include/asm-x86_64/calgary.h [new file with mode: 0644]
include/asm-x86_64/pci.h
include/asm-x86_64/tce.h [new file with mode: 0644]

index f2cd6ef53ff33e92f6545a43a7f2498c0582b107..6887d44d266188c9d5d17fc267b2441cc82d8e39 100644 (file)
@@ -205,6 +205,27 @@ IOMMU
   pages  Prereserve that many 128K pages for the software IO bounce buffering.
   force  Force all IO through the software TLB.
 
+  calgary=[64k,128k,256k,512k,1M,2M,4M,8M]
+  calgary=[translate_empty_slots]
+  calgary=[disable=<PCI bus number>]
+
+    64k,...,8M - Set the size of each PCI slot's translation table
+    when using the Calgary IOMMU. This is the size of the translation
+    table itself in main memory. The smallest table, 64k, covers an IO
+    space of 32MB; the largest, 8MB table, can cover an IO space of
+    4GB. Normally the kernel will make the right choice by itself.
+
+    translate_empty_slots - Enable translation even on slots that have
+    no devices attached to them, in case a device will be hotplugged
+    in the future.
+
+    disable=<PCI bus number> - Disable translation on a given PHB. For
+    example, the built-in graphics adapter resides on the first bridge
+    (PCI bus number 0); if translation (isolation) is enabled on this
+    bridge, X servers that access the hardware directly from user
+    space might stop working. Use this option if you have devices that
+    are accessed from userspace directly on some PCI host bridge.
+
 Debugging
 
   oops=panic Always panic on oopses. Default is to just kill the process,
index e8c52a1eec06505c736a456d8efa6d3ae4cff853..ccc4a7fb97a370f8d56baf16e4eb1439a17621d3 100644 (file)
@@ -405,6 +405,25 @@ config IOMMU
          device) unless CONFIG_IOMMU_DEBUG or iommu=force is specified
          too.
 
+config CALGARY_IOMMU
+       bool "IBM Calgary IOMMU support"
+       default y
+       select SWIOTLB
+       depends on PCI && EXPERIMENTAL
+       help
+         Support for hardware IOMMUs in IBM's xSeries x366 and x460
+         systems. Needed to run systems with more than 3GB of memory
+         properly with 32-bit PCI devices that do not support DAC
+         (Double Address Cycle). Calgary also supports bus level
+         isolation, where all DMAs pass through the IOMMU.  This
+         prevents them from going anywhere except their intended
+         destination. This catches hard-to-find kernel bugs and
+         mis-behaving drivers and devices that do not use the DMA-API
+         properly to set up their DMA buffers.  The IOMMU can be
+         turned off at boot time with the iommu=off parameter.
+         Normally the kernel will make the right choice by itself.
+         If unsure, say Y.
+
 # need this always selected by IOMMU for the VIA workaround
 config SWIOTLB
        bool
index fd106bdddd6d0104816e282a9fb81ee266411b85..aeb9c560be88aef2fd7141c792221ed85f92266e 100644 (file)
@@ -29,6 +29,7 @@ obj-$(CONFIG_SOFTWARE_SUSPEND)        += suspend_asm.o
 obj-$(CONFIG_CPU_FREQ)         += cpufreq/
 obj-$(CONFIG_EARLY_PRINTK)     += early_printk.o
 obj-$(CONFIG_IOMMU)            += pci-gart.o aperture.o
+obj-$(CONFIG_CALGARY_IOMMU)    += pci-calgary.o tce.o
 obj-$(CONFIG_SWIOTLB)          += pci-swiotlb.o
 obj-$(CONFIG_KPROBES)          += kprobes.o
 obj-$(CONFIG_X86_PM_TIMER)     += pmtimer.o
diff --git a/arch/x86_64/kernel/pci-calgary.c b/arch/x86_64/kernel/pci-calgary.c
new file mode 100644 (file)
index 0000000..d91cb84
--- /dev/null
@@ -0,0 +1,1018 @@
+/*
+ * Derived from arch/powerpc/kernel/iommu.c
+ *
+ * Copyright (C) 2006 Jon Mason <jdmason@us.ibm.com>, IBM Corporation
+ * Copyright (C) 2006 Muli Ben-Yehuda <muli@il.ibm.com>, IBM Corporation
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307 USA
+ */
+
+#include <linux/config.h>
+#include <linux/kernel.h>
+#include <linux/init.h>
+#include <linux/types.h>
+#include <linux/slab.h>
+#include <linux/mm.h>
+#include <linux/spinlock.h>
+#include <linux/string.h>
+#include <linux/dma-mapping.h>
+#include <linux/init.h>
+#include <linux/bitops.h>
+#include <linux/pci_ids.h>
+#include <linux/pci.h>
+#include <linux/delay.h>
+#include <asm/proto.h>
+#include <asm/calgary.h>
+#include <asm/tce.h>
+#include <asm/pci-direct.h>
+#include <asm/system.h>
+#include <asm/dma.h>
+
+#define PCI_DEVICE_ID_IBM_CALGARY 0x02a1
+#define PCI_VENDOR_DEVICE_ID_CALGARY \
+       (PCI_VENDOR_ID_IBM | PCI_DEVICE_ID_IBM_CALGARY << 16)
+
+/* we need these for register space address calculation */
+#define START_ADDRESS           0xfe000000
+#define CHASSIS_BASE            0
+#define ONE_BASED_CHASSIS_NUM   1
+
+/* register offsets inside the host bridge space */
+#define PHB_CSR_OFFSET         0x0110
+#define PHB_PLSSR_OFFSET       0x0120
+#define PHB_CONFIG_RW_OFFSET   0x0160
+#define PHB_IOBASE_BAR_LOW     0x0170
+#define PHB_IOBASE_BAR_HIGH    0x0180
+#define PHB_MEM_1_LOW          0x0190
+#define PHB_MEM_1_HIGH         0x01A0
+#define PHB_IO_ADDR_SIZE       0x01B0
+#define PHB_MEM_1_SIZE         0x01C0
+#define PHB_MEM_ST_OFFSET      0x01D0
+#define PHB_AER_OFFSET         0x0200
+#define PHB_CONFIG_0_HIGH      0x0220
+#define PHB_CONFIG_0_LOW       0x0230
+#define PHB_CONFIG_0_END       0x0240
+#define PHB_MEM_2_LOW          0x02B0
+#define PHB_MEM_2_HIGH         0x02C0
+#define PHB_MEM_2_SIZE_HIGH    0x02D0
+#define PHB_MEM_2_SIZE_LOW     0x02E0
+#define PHB_DOSHOLE_OFFSET     0x08E0
+
+/* PHB_CONFIG_RW */
+#define PHB_TCE_ENABLE         0x20000000
+#define PHB_SLOT_DISABLE       0x1C000000
+#define PHB_DAC_DISABLE                0x01000000
+#define PHB_MEM2_ENABLE                0x00400000
+#define PHB_MCSR_ENABLE                0x00100000
+/* TAR (Table Address Register) */
+#define TAR_SW_BITS            0x0000ffffffff800fUL
+#define TAR_VALID              0x0000000000000008UL
+/* CSR (Channel/DMA Status Register) */
+#define CSR_AGENT_MASK         0xffe0ffff
+
+#define MAX_NUM_OF_PHBS                8 /* how many PHBs in total? */
+#define MAX_PHB_BUS_NUM                (MAX_NUM_OF_PHBS * 2) /* max dev->bus->number */
+#define PHBS_PER_CALGARY       4
+
+/* register offsets in Calgary's internal register space */
+static const unsigned long tar_offsets[] = {
+       0x0580 /* TAR0 */,
+       0x0588 /* TAR1 */,
+       0x0590 /* TAR2 */,
+       0x0598 /* TAR3 */
+};
+
+static const unsigned long split_queue_offsets[] = {
+       0x4870 /* SPLIT QUEUE 0 */,
+       0x5870 /* SPLIT QUEUE 1 */,
+       0x6870 /* SPLIT QUEUE 2 */,
+       0x7870 /* SPLIT QUEUE 3 */
+};
+
+static const unsigned long phb_offsets[] = {
+       0x8000 /* PHB0 */,
+       0x9000 /* PHB1 */,
+       0xA000 /* PHB2 */,
+       0xB000 /* PHB3 */
+};
+
+void* tce_table_kva[MAX_NUM_OF_PHBS * MAX_NUMNODES];
+unsigned int specified_table_size = TCE_TABLE_SIZE_UNSPECIFIED;
+static int translate_empty_slots __read_mostly = 0;
+static int calgary_detected __read_mostly = 0;
+
+/*
+ * the bitmap of PHBs the user requested that we disable
+ * translation on.
+ */
+static DECLARE_BITMAP(translation_disabled, MAX_NUMNODES * MAX_PHB_BUS_NUM);
+
+static void tce_cache_blast(struct iommu_table *tbl);
+
+/* enable this to stress test the chip's TCE cache */
+#ifdef CONFIG_IOMMU_DEBUG
+static inline void tce_cache_blast_stress(struct iommu_table *tbl)
+{
+       tce_cache_blast(tbl);
+}
+#else
+static inline void tce_cache_blast_stress(struct iommu_table *tbl)
+{
+}
+#endif /* BLAST_TCE_CACHE_ON_UNMAP */
+
+static inline unsigned int num_dma_pages(unsigned long dma, unsigned int dmalen)
+{
+       unsigned int npages;
+
+       npages = PAGE_ALIGN(dma + dmalen) - (dma & PAGE_MASK);
+       npages >>= PAGE_SHIFT;
+
+       return npages;
+}
+
+static inline int translate_phb(struct pci_dev* dev)
+{
+       int disabled = test_bit(dev->bus->number, translation_disabled);
+       return !disabled;
+}
+
+static void iommu_range_reserve(struct iommu_table *tbl,
+        unsigned long start_addr, unsigned int npages)
+{
+       unsigned long index;
+       unsigned long end;
+
+       index = start_addr >> PAGE_SHIFT;
+
+       /* bail out if we're asked to reserve a region we don't cover */
+       if (index >= tbl->it_size)
+               return;
+
+       end = index + npages;
+       if (end > tbl->it_size) /* don't go off the table */
+               end = tbl->it_size;
+
+       while (index < end) {
+               if (test_bit(index, tbl->it_map))
+                       printk(KERN_ERR "Calgary: entry already allocated at "
+                              "0x%lx tbl %p dma 0x%lx npages %u\n",
+                              index, tbl, start_addr, npages);
+               ++index;
+       }
+       set_bit_string(tbl->it_map, start_addr >> PAGE_SHIFT, npages);
+}
+
+static unsigned long iommu_range_alloc(struct iommu_table *tbl,
+       unsigned int npages)
+{
+       unsigned long offset;
+
+       BUG_ON(npages == 0);
+
+       offset = find_next_zero_string(tbl->it_map, tbl->it_hint,
+                                      tbl->it_size, npages);
+       if (offset == ~0UL) {
+               tce_cache_blast(tbl);
+               offset = find_next_zero_string(tbl->it_map, 0,
+                                              tbl->it_size, npages);
+               if (offset == ~0UL) {
+                       printk(KERN_WARNING "Calgary: IOMMU full.\n");
+                       if (panic_on_overflow)
+                               panic("Calgary: fix the allocator.\n");
+                       else
+                               return bad_dma_address;
+               }
+       }
+
+       set_bit_string(tbl->it_map, offset, npages);
+       tbl->it_hint = offset + npages;
+       BUG_ON(tbl->it_hint > tbl->it_size);
+
+       return offset;
+}
+
+static dma_addr_t iommu_alloc(struct iommu_table *tbl, void *vaddr,
+       unsigned int npages, int direction)
+{
+       unsigned long entry, flags;
+       dma_addr_t ret = bad_dma_address;
+
+       spin_lock_irqsave(&tbl->it_lock, flags);
+
+       entry = iommu_range_alloc(tbl, npages);
+
+       if (unlikely(entry == bad_dma_address))
+               goto error;
+
+       /* set the return dma address */
+       ret = (entry << PAGE_SHIFT) | ((unsigned long)vaddr & ~PAGE_MASK);
+
+       /* put the TCEs in the HW table */
+       tce_build(tbl, entry, npages, (unsigned long)vaddr & PAGE_MASK,
+                 direction);
+
+       spin_unlock_irqrestore(&tbl->it_lock, flags);
+
+       return ret;
+
+error:
+       spin_unlock_irqrestore(&tbl->it_lock, flags);
+       printk(KERN_WARNING "Calgary: failed to allocate %u pages in "
+              "iommu %p\n", npages, tbl);
+       return bad_dma_address;
+}
+
+static void __iommu_free(struct iommu_table *tbl, dma_addr_t dma_addr,
+       unsigned int npages)
+{
+       unsigned long entry;
+       unsigned long i;
+
+       entry = dma_addr >> PAGE_SHIFT;
+
+       BUG_ON(entry + npages > tbl->it_size);
+
+       tce_free(tbl, entry, npages);
+
+       for (i = 0; i < npages; ++i) {
+               if (!test_bit(entry + i, tbl->it_map))
+                       printk(KERN_ERR "Calgary: bit is off at 0x%lx "
+                              "tbl %p dma 0x%Lx entry 0x%lx npages %u\n",
+                              entry + i, tbl, dma_addr, entry, npages);
+       }
+
+       __clear_bit_string(tbl->it_map, entry, npages);
+
+       tce_cache_blast_stress(tbl);
+}
+
+static void iommu_free(struct iommu_table *tbl, dma_addr_t dma_addr,
+       unsigned int npages)
+{
+       unsigned long flags;
+
+       spin_lock_irqsave(&tbl->it_lock, flags);
+
+       __iommu_free(tbl, dma_addr, npages);
+
+       spin_unlock_irqrestore(&tbl->it_lock, flags);
+}
+
+static void __calgary_unmap_sg(struct iommu_table *tbl,
+       struct scatterlist *sglist, int nelems, int direction)
+{
+       while (nelems--) {
+               unsigned int npages;
+               dma_addr_t dma = sglist->dma_address;
+               unsigned int dmalen = sglist->dma_length;
+
+               if (dmalen == 0)
+                       break;
+
+               npages = num_dma_pages(dma, dmalen);
+               __iommu_free(tbl, dma, npages);
+               sglist++;
+       }
+}
+
+void calgary_unmap_sg(struct device *dev, struct scatterlist *sglist,
+                     int nelems, int direction)
+{
+       unsigned long flags;
+       struct iommu_table *tbl = to_pci_dev(dev)->bus->self->sysdata;
+
+       if (!translate_phb(to_pci_dev(dev)))
+               return;
+
+       spin_lock_irqsave(&tbl->it_lock, flags);
+
+       __calgary_unmap_sg(tbl, sglist, nelems, direction);
+
+       spin_unlock_irqrestore(&tbl->it_lock, flags);
+}
+
+static int calgary_nontranslate_map_sg(struct device* dev,
+       struct scatterlist *sg, int nelems, int direction)
+{
+       int i;
+
+       for (i = 0; i < nelems; i++ ) {
+               struct scatterlist *s = &sg[i];
+               BUG_ON(!s->page);
+               s->dma_address = virt_to_bus(page_address(s->page) +s->offset);
+               s->dma_length = s->length;
+       }
+       return nelems;
+}
+
+int calgary_map_sg(struct device *dev, struct scatterlist *sg,
+       int nelems, int direction)
+{
+       struct iommu_table *tbl = to_pci_dev(dev)->bus->self->sysdata;
+       unsigned long flags;
+       unsigned long vaddr;
+       unsigned int npages;
+       unsigned long entry;
+       int i;
+
+       if (!translate_phb(to_pci_dev(dev)))
+               return calgary_nontranslate_map_sg(dev, sg, nelems, direction);
+
+       spin_lock_irqsave(&tbl->it_lock, flags);
+
+       for (i = 0; i < nelems; i++ ) {
+               struct scatterlist *s = &sg[i];
+               BUG_ON(!s->page);
+
+               vaddr = (unsigned long)page_address(s->page) + s->offset;
+               npages = num_dma_pages(vaddr, s->length);
+
+               entry = iommu_range_alloc(tbl, npages);
+               if (entry == bad_dma_address) {
+                       /* makes sure unmap knows to stop */
+                       s->dma_length = 0;
+                       goto error;
+               }
+
+               s->dma_address = (entry << PAGE_SHIFT) | s->offset;
+
+               /* insert into HW table */
+               tce_build(tbl, entry, npages, vaddr & PAGE_MASK,
+                         direction);
+
+               s->dma_length = s->length;
+       }
+
+       spin_unlock_irqrestore(&tbl->it_lock, flags);
+
+       return nelems;
+error:
+       __calgary_unmap_sg(tbl, sg, nelems, direction);
+       for (i = 0; i < nelems; i++) {
+               sg[i].dma_address = bad_dma_address;
+               sg[i].dma_length = 0;
+       }
+       spin_unlock_irqrestore(&tbl->it_lock, flags);
+       return 0;
+}
+
+dma_addr_t calgary_map_single(struct device *dev, void *vaddr,
+       size_t size, int direction)
+{
+       dma_addr_t dma_handle = bad_dma_address;
+       unsigned long uaddr;
+       unsigned int npages;
+       struct iommu_table *tbl = to_pci_dev(dev)->bus->self->sysdata;
+
+       uaddr = (unsigned long)vaddr;
+       npages = num_dma_pages(uaddr, size);
+
+       if (translate_phb(to_pci_dev(dev)))
+               dma_handle = iommu_alloc(tbl, vaddr, npages, direction);
+       else
+               dma_handle = virt_to_bus(vaddr);
+
+       return dma_handle;
+}
+
+void calgary_unmap_single(struct device *dev, dma_addr_t dma_handle,
+       size_t size, int direction)
+{
+       struct iommu_table *tbl = to_pci_dev(dev)->bus->self->sysdata;
+       unsigned int npages;
+
+       if (!translate_phb(to_pci_dev(dev)))
+               return;
+
+       npages = num_dma_pages(dma_handle, size);
+       iommu_free(tbl, dma_handle, npages);
+}
+
+void* calgary_alloc_coherent(struct device *dev, size_t size,
+       dma_addr_t *dma_handle, gfp_t flag)
+{
+       void *ret = NULL;
+       dma_addr_t mapping;
+       unsigned int npages, order;
+       struct iommu_table *tbl;
+
+       tbl = to_pci_dev(dev)->bus->self->sysdata;
+
+       size = PAGE_ALIGN(size); /* size rounded up to full pages */
+       npages = size >> PAGE_SHIFT;
+       order = get_order(size);
+
+       /* alloc enough pages (and possibly more) */
+       ret = (void *)__get_free_pages(flag, order);
+       if (!ret)
+               goto error;
+       memset(ret, 0, size);
+
+       if (translate_phb(to_pci_dev(dev))) {
+               /* set up tces to cover the allocated range */
+               mapping = iommu_alloc(tbl, ret, npages, DMA_BIDIRECTIONAL);
+               if (mapping == bad_dma_address)
+                       goto free;
+
+               *dma_handle = mapping;
+       } else /* non translated slot */
+               *dma_handle = virt_to_bus(ret);
+
+       return ret;
+
+free:
+       free_pages((unsigned long)ret, get_order(size));
+       ret = NULL;
+error:
+       return ret;
+}
+
+static struct dma_mapping_ops calgary_dma_ops = {
+       .alloc_coherent = calgary_alloc_coherent,
+       .map_single = calgary_map_single,
+       .unmap_single = calgary_unmap_single,
+       .map_sg = calgary_map_sg,
+       .unmap_sg = calgary_unmap_sg,
+};
+
+static inline int busno_to_phbid(unsigned char num)
+{
+       return bus_to_phb(num) % PHBS_PER_CALGARY;
+}
+
+static inline unsigned long split_queue_offset(unsigned char num)
+{
+       size_t idx = busno_to_phbid(num);
+
+       return split_queue_offsets[idx];
+}
+
+static inline unsigned long tar_offset(unsigned char num)
+{
+       size_t idx = busno_to_phbid(num);
+
+       return tar_offsets[idx];
+}
+
+static inline unsigned long phb_offset(unsigned char num)
+{
+       size_t idx = busno_to_phbid(num);
+
+       return phb_offsets[idx];
+}
+
+static inline void __iomem* calgary_reg(void __iomem *bar, unsigned long offset)
+{
+       unsigned long target = ((unsigned long)bar) | offset;
+       return (void __iomem*)target;
+}
+
+static void tce_cache_blast(struct iommu_table *tbl)
+{
+       u64 val;
+       u32 aer;
+       int i = 0;
+       void __iomem *bbar = tbl->bbar;
+       void __iomem *target;
+
+       /* disable arbitration on the bus */
+       target = calgary_reg(bbar, phb_offset(tbl->it_busno) | PHB_AER_OFFSET);
+       aer = readl(target);
+       writel(0, target);
+
+       /* read plssr to ensure it got there */
+       target = calgary_reg(bbar, phb_offset(tbl->it_busno) | PHB_PLSSR_OFFSET);
+       val = readl(target);
+
+       /* poll split queues until all DMA activity is done */
+       target = calgary_reg(bbar, split_queue_offset(tbl->it_busno));
+       do {
+               val = readq(target);
+               i++;
+       } while ((val & 0xff) != 0xff && i < 100);
+       if (i == 100)
+               printk(KERN_WARNING "Calgary: PCI bus not quiesced, "
+                      "continuing anyway\n");
+
+       /* invalidate TCE cache */
+       target = calgary_reg(bbar, tar_offset(tbl->it_busno));
+       writeq(tbl->tar_val, target);
+
+       /* enable arbitration */
+       target = calgary_reg(bbar, phb_offset(tbl->it_busno) | PHB_AER_OFFSET);
+       writel(aer, target);
+       (void)readl(target); /* flush */
+}
+
+static void __init calgary_reserve_mem_region(struct pci_dev *dev, u64 start,
+       u64 limit)
+{
+       unsigned int numpages;
+
+       limit = limit | 0xfffff;
+       limit++;
+
+       numpages = ((limit - start) >> PAGE_SHIFT);
+       iommu_range_reserve(dev->sysdata, start, numpages);
+}
+
+static void __init calgary_reserve_peripheral_mem_1(struct pci_dev *dev)
+{
+       void __iomem *target;
+       u64 low, high, sizelow;
+       u64 start, limit;
+       struct iommu_table *tbl = dev->sysdata;
+       unsigned char busnum = dev->bus->number;
+       void __iomem *bbar = tbl->bbar;
+
+       /* peripheral MEM_1 region */
+       target = calgary_reg(bbar, phb_offset(busnum) | PHB_MEM_1_LOW);
+       low = be32_to_cpu(readl(target));
+       target = calgary_reg(bbar, phb_offset(busnum) | PHB_MEM_1_HIGH);
+       high = be32_to_cpu(readl(target));
+       target = calgary_reg(bbar, phb_offset(busnum) | PHB_MEM_1_SIZE);
+       sizelow = be32_to_cpu(readl(target));
+
+       start = (high << 32) | low;
+       limit = sizelow;
+
+       calgary_reserve_mem_region(dev, start, limit);
+}
+
+static void __init calgary_reserve_peripheral_mem_2(struct pci_dev *dev)
+{
+       void __iomem *target;
+       u32 val32;
+       u64 low, high, sizelow, sizehigh;
+       u64 start, limit;
+       struct iommu_table *tbl = dev->sysdata;
+       unsigned char busnum = dev->bus->number;
+       void __iomem *bbar = tbl->bbar;
+
+       /* is it enabled? */
+       target = calgary_reg(bbar, phb_offset(busnum) | PHB_CONFIG_RW_OFFSET);
+       val32 = be32_to_cpu(readl(target));
+       if (!(val32 & PHB_MEM2_ENABLE))
+               return;
+
+       target = calgary_reg(bbar, phb_offset(busnum) | PHB_MEM_2_LOW);
+       low = be32_to_cpu(readl(target));
+       target = calgary_reg(bbar, phb_offset(busnum) | PHB_MEM_2_HIGH);
+       high = be32_to_cpu(readl(target));
+       target = calgary_reg(bbar, phb_offset(busnum) | PHB_MEM_2_SIZE_LOW);
+       sizelow = be32_to_cpu(readl(target));
+       target = calgary_reg(bbar, phb_offset(busnum) | PHB_MEM_2_SIZE_HIGH);
+       sizehigh = be32_to_cpu(readl(target));
+
+       start = (high << 32) | low;
+       limit = (sizehigh << 32) | sizelow;
+
+       calgary_reserve_mem_region(dev, start, limit);
+}
+
+/*
+ * some regions of the IO address space do not get translated, so we
+ * must not give devices IO addresses in those regions. The regions
+ * are the 640KB-1MB region and the two PCI peripheral memory holes.
+ * Reserve all of them in the IOMMU bitmap to avoid giving them out
+ * later.
+ */
+static void __init calgary_reserve_regions(struct pci_dev *dev)
+{
+       unsigned int npages;
+       void __iomem *bbar;
+       unsigned char busnum;
+       u64 start;
+       struct iommu_table *tbl = dev->sysdata;
+
+       bbar = tbl->bbar;
+       busnum = dev->bus->number;
+
+       /* reserve bad_dma_address in case it's a legal address */
+       iommu_range_reserve(tbl, bad_dma_address, 1);
+
+       /* avoid the BIOS/VGA first 640KB-1MB region */
+       start = (640 * 1024);
+       npages = ((1024 - 640) * 1024) >> PAGE_SHIFT;
+       iommu_range_reserve(tbl, start, npages);
+
+       /* reserve the two PCI peripheral memory regions in IO space */
+       calgary_reserve_peripheral_mem_1(dev);
+       calgary_reserve_peripheral_mem_2(dev);
+}
+
+static int __init calgary_setup_tar(struct pci_dev *dev, void __iomem *bbar)
+{
+       u64 val64;
+       u64 table_phys;
+       void __iomem *target;
+       int ret;
+       struct iommu_table *tbl;
+
+       /* build TCE tables for each PHB */
+       ret = build_tce_table(dev, bbar);
+       if (ret)
+               return ret;
+
+       calgary_reserve_regions(dev);
+
+       /* set TARs for each PHB */
+       target = calgary_reg(bbar, tar_offset(dev->bus->number));
+       val64 = be64_to_cpu(readq(target));
+
+       /* zero out all TAR bits under sw control */
+       val64 &= ~TAR_SW_BITS;
+
+       tbl = dev->sysdata;
+       table_phys = (u64)__pa(tbl->it_base);
+       val64 |= table_phys;
+
+       BUG_ON(specified_table_size > TCE_TABLE_SIZE_8M);
+       val64 |= (u64) specified_table_size;
+
+       tbl->tar_val = cpu_to_be64(val64);
+       writeq(tbl->tar_val, target);
+       readq(target); /* flush */
+
+       return 0;
+}
+
+static void __init calgary_free_tar(struct pci_dev *dev)
+{
+       u64 val64;
+       struct iommu_table *tbl = dev->sysdata;
+       void __iomem *target;
+
+       target = calgary_reg(tbl->bbar, tar_offset(dev->bus->number));
+       val64 = be64_to_cpu(readq(target));
+       val64 &= ~TAR_SW_BITS;
+       writeq(cpu_to_be64(val64), target);
+       readq(target); /* flush */
+
+       kfree(tbl);
+       dev->sysdata = NULL;
+}
+
+static void calgary_watchdog(unsigned long data)
+{
+       struct pci_dev *dev = (struct pci_dev *)data;
+       struct iommu_table *tbl = dev->sysdata;
+       void __iomem *bbar = tbl->bbar;
+       u32 val32;
+       void __iomem *target;
+
+       target = calgary_reg(bbar, phb_offset(tbl->it_busno) | PHB_CSR_OFFSET);
+       val32 = be32_to_cpu(readl(target));
+
+       /* If no error, the agent ID in the CSR is not valid */
+       if (val32 & CSR_AGENT_MASK) {
+               printk(KERN_EMERG "calgary_watchdog: DMA error on bus %d, "
+                                 "CSR = %#x\n", dev->bus->number, val32);
+               writel(0, target);
+
+               /* Disable bus that caused the error */
+               target = calgary_reg(bbar, phb_offset(tbl->it_busno) |
+                                          PHB_CONFIG_RW_OFFSET);
+               val32 = be32_to_cpu(readl(target));
+               val32 |= PHB_SLOT_DISABLE;
+               writel(cpu_to_be32(val32), target);
+               readl(target); /* flush */
+       } else {
+               /* Reset the timer */
+               mod_timer(&tbl->watchdog_timer, jiffies + 2 * HZ);
+       }
+}
+
+static void __init calgary_enable_translation(struct pci_dev *dev)
+{
+       u32 val32;
+       unsigned char busnum;
+       void __iomem *target;
+       void __iomem *bbar;
+       struct iommu_table *tbl;
+
+       busnum = dev->bus->number;
+       tbl = dev->sysdata;
+       bbar = tbl->bbar;
+
+       /* enable TCE in PHB Config Register */
+       target = calgary_reg(bbar, phb_offset(busnum) | PHB_CONFIG_RW_OFFSET);
+       val32 = be32_to_cpu(readl(target));
+       val32 |= PHB_TCE_ENABLE | PHB_DAC_DISABLE | PHB_MCSR_ENABLE;
+
+       printk(KERN_INFO "Calgary: enabling translation on PHB %d\n", busnum);
+       printk(KERN_INFO "Calgary: errant DMAs will now be prevented on this "
+              "bus.\n");
+
+       writel(cpu_to_be32(val32), target);
+       readl(target); /* flush */
+
+       init_timer(&tbl->watchdog_timer);
+       tbl->watchdog_timer.function = &calgary_watchdog;
+       tbl->watchdog_timer.data = (unsigned long)dev;
+       mod_timer(&tbl->watchdog_timer, jiffies);
+}
+
+static void __init calgary_disable_translation(struct pci_dev *dev)
+{
+       u32 val32;
+       unsigned char busnum;
+       void __iomem *target;
+       void __iomem *bbar;
+       struct iommu_table *tbl;
+
+       busnum = dev->bus->number;
+       tbl = dev->sysdata;
+       bbar = tbl->bbar;
+
+       /* disable TCE in PHB Config Register */
+       target = calgary_reg(bbar, phb_offset(busnum) | PHB_CONFIG_RW_OFFSET);
+       val32 = be32_to_cpu(readl(target));
+       val32 &= ~(PHB_TCE_ENABLE | PHB_DAC_DISABLE | PHB_MCSR_ENABLE);
+
+       printk(KERN_INFO "Calgary: disabling translation on PHB %d!\n", busnum);
+       writel(cpu_to_be32(val32), target);
+       readl(target); /* flush */
+
+       del_timer_sync(&tbl->watchdog_timer);
+}
+
+static inline unsigned int __init locate_register_space(struct pci_dev *dev)
+{
+       int rionodeid;
+       u32 address;
+
+       rionodeid = (dev->bus->number % 15 > 4) ? 3 : 2;
+       /*
+        * register space address calculation as follows:
+        * FE0MB-8MB*OneBasedChassisNumber+1MB*(RioNodeId-ChassisBase)
+        * ChassisBase is always zero for x366/x260/x460
+        * RioNodeId is 2 for first Calgary, 3 for second Calgary
+        */
+       address = START_ADDRESS -
+               (0x800000 * (ONE_BASED_CHASSIS_NUM + dev->bus->number / 15)) +
+               (0x100000) * (rionodeid - CHASSIS_BASE);
+       return address;
+}
+
+static int __init calgary_init_one_nontraslated(struct pci_dev *dev)
+{
+       dev->sysdata = NULL;
+       dev->bus->self = dev;
+
+       return 0;
+}
+
+static int __init calgary_init_one(struct pci_dev *dev)
+{
+       u32 address;
+       void __iomem *bbar;
+       int ret;
+
+       address = locate_register_space(dev);
+       /* map entire 1MB of Calgary config space */
+       bbar = ioremap_nocache(address, 1024 * 1024);
+       if (!bbar) {
+               ret = -ENODATA;
+               goto done;
+       }
+
+       ret = calgary_setup_tar(dev, bbar);
+       if (ret)
+               goto iounmap;
+
+       dev->bus->self = dev;
+       calgary_enable_translation(dev);
+
+       return 0;
+
+iounmap:
+       iounmap(bbar);
+done:
+       return ret;
+}
+
+static int __init calgary_init(void)
+{
+       int i, ret = -ENODEV;
+       struct pci_dev *dev = NULL;
+
+       for (i = 0; i <= num_online_nodes() * MAX_NUM_OF_PHBS; i++) {
+               dev = pci_get_device(PCI_VENDOR_ID_IBM,
+                                    PCI_DEVICE_ID_IBM_CALGARY,
+                                    dev);
+               if (!dev)
+                       break;
+               if (!translate_phb(dev)) {
+                       calgary_init_one_nontraslated(dev);
+                       continue;
+               }
+               if (!tce_table_kva[i] && !translate_empty_slots) {
+                       pci_dev_put(dev);
+                       continue;
+               }
+               ret = calgary_init_one(dev);
+               if (ret)
+                       goto error;
+       }
+
+       return ret;
+
+error:
+       for (i--; i >= 0; i--) {
+               dev = pci_find_device_reverse(PCI_VENDOR_ID_IBM,
+                                             PCI_DEVICE_ID_IBM_CALGARY,
+                                             dev);
+               if (!translate_phb(dev)) {
+                       pci_dev_put(dev);
+                       continue;
+               }
+               if (!tce_table_kva[i] && !translate_empty_slots)
+                       continue;
+               calgary_disable_translation(dev);
+               calgary_free_tar(dev);
+               pci_dev_put(dev);
+       }
+
+       return ret;
+}
+
+static inline int __init determine_tce_table_size(u64 ram)
+{
+       int ret;
+
+       if (specified_table_size != TCE_TABLE_SIZE_UNSPECIFIED)
+               return specified_table_size;
+
+       /*
+        * Table sizes are from 0 to 7 (TCE_TABLE_SIZE_64K to
+        * TCE_TABLE_SIZE_8M). Table size 0 has 8K entries and each
+        * larger table size has twice as many entries, so shift the
+        * max ram address by 13 to divide by 8K and then look at the
+        * order of the result to choose between 0-7.
+        */
+       ret = get_order(ram >> 13);
+       if (ret > TCE_TABLE_SIZE_8M)
+               ret = TCE_TABLE_SIZE_8M;
+
+       return ret;
+}
+
+void __init detect_calgary(void)
+{
+       u32 val;
+       int bus, table_idx;
+       void *tbl;
+       int detected = 0;
+
+       /*
+        * if the user specified iommu=off or iommu=soft or we found
+        * another HW IOMMU already, bail out.
+        */
+       if (swiotlb || no_iommu || iommu_detected)
+               return;
+
+       specified_table_size = determine_tce_table_size(end_pfn * PAGE_SIZE);
+
+       for (bus = 0, table_idx = 0;
+            bus <= num_online_nodes() * MAX_PHB_BUS_NUM;
+            bus++) {
+               BUG_ON(bus > MAX_NUMNODES * MAX_PHB_BUS_NUM);
+               if (read_pci_config(bus, 0, 0, 0) != PCI_VENDOR_DEVICE_ID_CALGARY)
+                       continue;
+               if (test_bit(bus, translation_disabled)) {
+                       printk(KERN_INFO "Calgary: translation is disabled for "
+                              "PHB 0x%x\n", bus);
+                       /* skip this phb, don't allocate a tbl for it */
+                       tce_table_kva[table_idx] = NULL;
+                       table_idx++;
+                       continue;
+               }
+               /*
+                * scan the first slot of the PCI bus to see if there
+                * are any devices present
+                */
+               val = read_pci_config(bus, 1, 0, 0);
+               if (val != 0xffffffff || translate_empty_slots) {
+                       tbl = alloc_tce_table();
+                       if (!tbl)
+                               goto cleanup;
+                       detected = 1;
+               } else
+                       tbl = NULL;
+
+               tce_table_kva[table_idx] = tbl;
+               table_idx++;
+       }
+
+       if (detected) {
+               iommu_detected = 1;
+               calgary_detected = 1;
+               printk(KERN_INFO "PCI-DMA: Calgary IOMMU detected. "
+                      "TCE table spec is %d.\n", specified_table_size);
+       }
+       return;
+
+cleanup:
+       for (--table_idx; table_idx >= 0; --table_idx)
+               if (tce_table_kva[table_idx])
+                       free_tce_table(tce_table_kva[table_idx]);
+}
+
+int __init calgary_iommu_init(void)
+{
+       int ret;
+
+       if (no_iommu || swiotlb)
+               return -ENODEV;
+
+       if (!calgary_detected)
+               return -ENODEV;
+
+       /* ok, we're trying to use Calgary - let's roll */
+       printk(KERN_INFO "PCI-DMA: Using Calgary IOMMU\n");
+
+       ret = calgary_init();
+       if (ret) {
+               printk(KERN_ERR "PCI-DMA: Calgary init failed %d, "
+                      "falling back to no_iommu\n", ret);
+               if (end_pfn > MAX_DMA32_PFN)
+                       printk(KERN_ERR "WARNING more than 4GB of memory, "
+                                       "32bit PCI may malfunction.\n");
+               return ret;
+       }
+
+       force_iommu = 1;
+       dma_ops = &calgary_dma_ops;
+
+       return 0;
+}
+
+static int __init calgary_parse_options(char *p)
+{
+       unsigned int bridge;
+       size_t len;
+       char* endp;
+
+       while (*p) {
+               if (!strncmp(p, "64k", 3))
+                       specified_table_size = TCE_TABLE_SIZE_64K;
+               else if (!strncmp(p, "128k", 4))
+                       specified_table_size = TCE_TABLE_SIZE_128K;
+               else if (!strncmp(p, "256k", 4))
+                       specified_table_size = TCE_TABLE_SIZE_256K;
+               else if (!strncmp(p, "512k", 4))
+                       specified_table_size = TCE_TABLE_SIZE_512K;
+               else if (!strncmp(p, "1M", 2))
+                       specified_table_size = TCE_TABLE_SIZE_1M;
+               else if (!strncmp(p, "2M", 2))
+                       specified_table_size = TCE_TABLE_SIZE_2M;
+               else if (!strncmp(p, "4M", 2))
+                       specified_table_size = TCE_TABLE_SIZE_4M;
+               else if (!strncmp(p, "8M", 2))
+                       specified_table_size = TCE_TABLE_SIZE_8M;
+
+               len = strlen("translate_empty_slots");
+               if (!strncmp(p, "translate_empty_slots", len))
+                       translate_empty_slots = 1;
+
+               len = strlen("disable");
+               if (!strncmp(p, "disable", len)) {
+                       p += len;
+                       if (*p == '=')
+                               ++p;
+                       if (*p == '\0')
+                               break;
+                       bridge = simple_strtol(p, &endp, 0);
+                       if (p == endp)
+                               break;
+
+                       if (bridge <= (num_online_nodes() * MAX_PHB_BUS_NUM)) {
+                               printk(KERN_INFO "Calgary: disabling "
+                                      "translation for PHB 0x%x\n", bridge);
+                               set_bit(bridge, translation_disabled);
+                       }
+               }
+
+               p = strpbrk(p, ",");
+               if (!p)
+                       break;
+
+               p++; /* skip ',' */
+       }
+       return 1;
+}
+__setup("calgary=", calgary_parse_options);
index a45844c7e3a32b31cc5d853920a1ce3e00bcf2a4..9c44f4f2433d7b4d77ed1ee9e2192f14b0149228 100644 (file)
@@ -9,6 +9,7 @@
 #include <linux/module.h>
 #include <asm/io.h>
 #include <asm/proto.h>
+#include <asm/calgary.h>
 
 int iommu_merge __read_mostly = 0;
 EXPORT_SYMBOL(iommu_merge);
@@ -291,6 +292,10 @@ void __init pci_iommu_alloc(void)
        iommu_hole_init();
 #endif
 
+#ifdef CONFIG_CALGARY_IOMMU
+       detect_calgary();
+#endif
+
 #ifdef CONFIG_SWIOTLB
        pci_swiotlb_init();
 #endif
@@ -298,6 +303,10 @@ void __init pci_iommu_alloc(void)
 
 static int __init pci_iommu_init(void)
 {
+#ifdef CONFIG_CALGARY_IOMMU
+       calgary_iommu_init();
+#endif
+
 #ifdef CONFIG_IOMMU
        gart_iommu_init();
 #endif
diff --git a/arch/x86_64/kernel/tce.c b/arch/x86_64/kernel/tce.c
new file mode 100644 (file)
index 0000000..8d4c67f
--- /dev/null
@@ -0,0 +1,202 @@
+/*
+ * Derived from arch/powerpc/platforms/pseries/iommu.c
+ *
+ * Copyright (C) 2006 Jon Mason <jdmason@us.ibm.com>, IBM Corporation
+ * Copyright (C) 2006 Muli Ben-Yehuda <muli@il.ibm.com>, IBM Corporation
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307 USA
+ */
+
+#include <linux/config.h>
+#include <linux/types.h>
+#include <linux/slab.h>
+#include <linux/mm.h>
+#include <linux/spinlock.h>
+#include <linux/string.h>
+#include <linux/pci.h>
+#include <linux/dma-mapping.h>
+#include <linux/bootmem.h>
+#include <asm/tce.h>
+#include <asm/calgary.h>
+#include <asm/proto.h>
+
+/* flush a tce at 'tceaddr' to main memory */
+static inline void flush_tce(void* tceaddr)
+{
+       /* a single tce can't cross a cache line */
+       if (cpu_has_clflush)
+               asm volatile("clflush (%0)" :: "r" (tceaddr));
+       else
+               asm volatile("wbinvd":::"memory");
+}
+
+void tce_build(struct iommu_table *tbl, unsigned long index,
+       unsigned int npages, unsigned long uaddr, int direction)
+{
+       u64* tp;
+       u64 t;
+       u64 rpn;
+
+       t = (1 << TCE_READ_SHIFT);
+       if (direction != DMA_TO_DEVICE)
+               t |= (1 << TCE_WRITE_SHIFT);
+
+       tp = ((u64*)tbl->it_base) + index;
+
+       while (npages--) {
+               rpn = (virt_to_bus((void*)uaddr)) >> PAGE_SHIFT;
+               t &= ~TCE_RPN_MASK;
+               t |= (rpn << TCE_RPN_SHIFT);
+
+               *tp = cpu_to_be64(t);
+               flush_tce(tp);
+
+               uaddr += PAGE_SIZE;
+               tp++;
+       }
+}
+
+void tce_free(struct iommu_table *tbl, long index, unsigned int npages)
+{
+       u64* tp;
+
+       tp  = ((u64*)tbl->it_base) + index;
+
+       while (npages--) {
+               *tp = cpu_to_be64(0);
+               flush_tce(tp);
+               tp++;
+       }
+}
+
+static inline unsigned int table_size_to_number_of_entries(unsigned char size)
+{
+       /*
+        * size is the order of the table, 0-7
+        * smallest table is 8K entries, so shift result by 13 to
+        * multiply by 8K
+        */
+       return (1 << size) << 13;
+}
+
+static int tce_table_setparms(struct pci_dev *dev, struct iommu_table *tbl)
+{
+       unsigned int bitmapsz;
+       unsigned int tce_table_index;
+       unsigned long bmppages;
+       int ret;
+
+       tbl->it_busno = dev->bus->number;
+
+       /* set the tce table size - measured in entries */
+       tbl->it_size = table_size_to_number_of_entries(specified_table_size);
+
+       tce_table_index = bus_to_phb(tbl->it_busno);
+       tbl->it_base = (unsigned long)tce_table_kva[tce_table_index];
+       if (!tbl->it_base) {
+               printk(KERN_ERR "Calgary: iommu_table_setparms: "
+                      "no table allocated?!\n");
+               ret = -ENOMEM;
+               goto done;
+       }
+
+       /*
+        * number of bytes needed for the bitmap size in number of
+        * entries; we need one bit per entry
+        */
+       bitmapsz = tbl->it_size / BITS_PER_BYTE;
+       bmppages = __get_free_pages(GFP_KERNEL, get_order(bitmapsz));
+       if (!bmppages) {
+               printk(KERN_ERR "Calgary: cannot allocate bitmap\n");
+               ret = -ENOMEM;
+               goto done;
+       }
+
+       tbl->it_map = (unsigned long*)bmppages;
+
+       memset(tbl->it_map, 0, bitmapsz);
+
+       tbl->it_hint = 0;
+
+       spin_lock_init(&tbl->it_lock);
+
+       return 0;
+
+done:
+       return ret;
+}
+
+int build_tce_table(struct pci_dev *dev, void __iomem *bbar)
+{
+       struct iommu_table *tbl;
+       int ret;
+
+       if (dev->sysdata) {
+               printk(KERN_ERR "Calgary: dev %p has sysdata %p\n",
+                      dev, dev->sysdata);
+               BUG();
+       }
+
+       tbl = kzalloc(sizeof(struct iommu_table), GFP_KERNEL);
+       if (!tbl) {
+               printk(KERN_ERR "Calgary: error allocating iommu_table\n");
+               ret = -ENOMEM;
+               goto done;
+       }
+
+       ret = tce_table_setparms(dev, tbl);
+       if (ret)
+               goto free_tbl;
+
+       tce_free(tbl, 0, tbl->it_size);
+
+       tbl->bbar = bbar;
+
+       /*
+        * NUMA is already using the bus's sysdata pointer, so we use
+        * the bus's pci_dev's sysdata instead.
+        */
+       dev->sysdata = tbl;
+
+       return 0;
+
+free_tbl:
+       kfree(tbl);
+done:
+       return ret;
+}
+
+void* alloc_tce_table(void)
+{
+       unsigned int size;
+
+       size = table_size_to_number_of_entries(specified_table_size);
+       size *= TCE_ENTRY_SIZE;
+
+       return __alloc_bootmem_low(size, size, 0);
+}
+
+void free_tce_table(void *tbl)
+{
+       unsigned int size;
+
+       if (!tbl)
+               return;
+
+       size = table_size_to_number_of_entries(specified_table_size);
+       size *= TCE_ENTRY_SIZE;
+
+       free_bootmem(__pa(tbl), size);
+}
diff --git a/include/asm-x86_64/calgary.h b/include/asm-x86_64/calgary.h
new file mode 100644 (file)
index 0000000..6e1654f
--- /dev/null
@@ -0,0 +1,66 @@
+/*
+ * Derived from include/asm-powerpc/iommu.h
+ *
+ * Copyright (C) 2006 Jon Mason <jdmason@us.ibm.com>, IBM Corporation
+ * Copyright (C) 2006 Muli Ben-Yehuda <muli@il.ibm.com>, IBM Corporation
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307 USA
+ */
+
+#ifndef _ASM_X86_64_CALGARY_H
+#define _ASM_X86_64_CALGARY_H
+
+#include <linux/config.h>
+#include <linux/spinlock.h>
+#include <linux/device.h>
+#include <linux/dma-mapping.h>
+#include <asm/types.h>
+
+struct iommu_table {
+       unsigned long  it_base;      /* mapped address of tce table */
+       unsigned long  it_hint;      /* Hint for next alloc */
+       unsigned long *it_map;       /* A simple allocation bitmap for now */
+       spinlock_t     it_lock;      /* Protects it_map */
+       unsigned int   it_size;      /* Size of iommu table in entries */
+       unsigned char  it_busno;     /* Bus number this table belongs to */
+       void __iomem  *bbar;
+       u64            tar_val;
+       struct timer_list watchdog_timer;
+};
+
+#define TCE_TABLE_SIZE_UNSPECIFIED     ~0
+#define TCE_TABLE_SIZE_64K             0
+#define TCE_TABLE_SIZE_128K            1
+#define TCE_TABLE_SIZE_256K            2
+#define TCE_TABLE_SIZE_512K            3
+#define TCE_TABLE_SIZE_1M              4
+#define TCE_TABLE_SIZE_2M              5
+#define TCE_TABLE_SIZE_4M              6
+#define TCE_TABLE_SIZE_8M              7
+
+#ifdef CONFIG_CALGARY_IOMMU
+extern int calgary_iommu_init(void);
+extern void detect_calgary(void);
+#else
+static inline int calgary_iommu_init(void) { return 1; }
+static inline void detect_calgary(void) { return; }
+#endif
+
+static inline unsigned int bus_to_phb(unsigned char busno)
+{
+       return ((busno % 15 == 0) ? 0 : busno / 2 + 1);
+}
+
+#endif /* _ASM_X86_64_CALGARY_H */
index 4dbc07c54f7ab4c5e7a6aea79fbf619bc0d16487..49c5e9280598729000a42e07483fa5f21da2b538 100644 (file)
@@ -52,7 +52,7 @@ extern int iommu_setup(char *opt);
  */
 #define PCI_DMA_BUS_IS_PHYS (dma_ops->is_phys)
 
-#ifdef CONFIG_IOMMU
+#if defined(CONFIG_IOMMU) || defined(CONFIG_CALGARY_IOMMU)
 
 /*
  * x86-64 always supports DAC, but sometimes it is useful to force
diff --git a/include/asm-x86_64/tce.h b/include/asm-x86_64/tce.h
new file mode 100644 (file)
index 0000000..ee51d31
--- /dev/null
@@ -0,0 +1,47 @@
+/*
+ * Copyright (C) 2006 Muli Ben-Yehuda <muli@il.ibm.com>, IBM Corporation
+ * Copyright (C) 2006 Jon Mason <jdmason@us.ibm.com>, IBM Corporation
+ *
+ * This file is derived from asm-powerpc/tce.h.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307 USA
+ */
+
+#ifndef _ASM_X86_64_TCE_H
+#define _ASM_X86_64_TCE_H
+
+extern void* tce_table_kva[];
+extern unsigned int specified_table_size;
+struct iommu_table;
+
+#define TCE_ENTRY_SIZE   8   /* in bytes */
+
+#define TCE_READ_SHIFT   0
+#define TCE_WRITE_SHIFT  1
+#define TCE_HUBID_SHIFT  2   /* unused */
+#define TCE_RSVD_SHIFT   8   /* unused */
+#define TCE_RPN_SHIFT    12
+#define TCE_UNUSED_SHIFT 48  /* unused */
+
+#define TCE_RPN_MASK     0x0000fffffffff000ULL
+
+extern void tce_build(struct iommu_table *tbl, unsigned long index,
+        unsigned int npages, unsigned long uaddr, int direction);
+extern void tce_free(struct iommu_table *tbl, long index, unsigned int npages);
+extern void* alloc_tce_table(void);
+extern void free_tce_table(void *tbl);
+extern int build_tce_table(struct pci_dev *dev, void __iomem *bbar);
+
+#endif /* _ASM_X86_64_TCE_H */