arch/powerpc/sysdev/dart_iommu.c

   1 /*
   2  * arch/powerpc/sysdev/dart_iommu.c
   3  *
   4  * Copyright (C) 2004 Olof Johansson <olof@lixom.net>, IBM Corporation
   5  * Copyright (C) 2005 Benjamin Herrenschmidt <benh@kernel.crashing.org>,
   6  *                    IBM Corporation
   7  *
   8  * Based on pSeries_iommu.c:
   9  * Copyright (C) 2001 Mike Corrigan & Dave Engebretsen, IBM Corporation
  10  * Copyright (C) 2004 Olof Johansson <olof@lixom.net>, IBM Corporation
  11  *
  12  * Dynamic DMA mapping support, Apple U3, U4 & IBM CPC925 "DART" iommu.
  13  *
  14  *
  15  * This program is free software; you can redistribute it and/or modify
  16  * it under the terms of the GNU General Public License as published by
  17  * the Free Software Foundation; either version 2 of the License, or
  18  * (at your option) any later version.
  19  *
  20  * This program is distributed in the hope that it will be useful,
  21  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  22  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  23  * GNU General Public License for more details.
  24  *
  25  * You should have received a copy of the GNU General Public License
  26  * along with this program; if not, write to the Free Software
  27  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307 USA
  28  */
  29
  30 #include <linux/config.h>
  31 #include <linux/init.h>
  32 #include <linux/types.h>
  33 #include <linux/slab.h>
  34 #include <linux/mm.h>
  35 #include <linux/spinlock.h>
  36 #include <linux/string.h>
  37 #include <linux/pci.h>
  38 #include <linux/dma-mapping.h>
  39 #include <linux/vmalloc.h>
  40 #include <asm/io.h>
  41 #include <asm/prom.h>
  42 #include <asm/iommu.h>
  43 #include <asm/pci-bridge.h>
  44 #include <asm/machdep.h>
  45 #include <asm/abs_addr.h>
  46 #include <asm/cacheflush.h>
  47 #include <asm/lmb.h>
  48 #include <asm/ppc-pci.h>
  49
  50 #include "dart.h"
  51
  52 extern int iommu_force_on;
  53
  54 /* Physical base address and size of the DART table */
  55 unsigned long dart_tablebase; /* exported to htab_initialize */
  56 static unsigned long dart_tablesize;
  57
  58 /* Virtual base address of the DART table */
  59 static u32 *dart_vbase;
  60
  61 /* Mapped base address for the dart */
  62 static unsigned int __iomem *dart;
  63
  64 /* Dummy val that entries are set to when unused */
  65 static unsigned int dart_emptyval;
  66
  67 static struct iommu_table iommu_table_dart;
  68 static int iommu_table_dart_inited;
  69 static int dart_dirty;
  70 static int dart_is_u4;
  71
  72 #define DBG(...)
  73
  74 static inline void dart_tlb_invalidate_all(void)
  75 {
  76         unsigned long l = 0;
  77         unsigned int reg, inv_bit;
  78         unsigned long limit;
  79
  80         DBG("dart: flush\n");
  81
  82         /* To invalidate the DART, set the DARTCNTL_FLUSHTLB bit in the
  83          * control register and wait for it to clear.
  84          *
  85          * Gotcha: Sometimes, the DART won't detect that the bit gets
  86          * set. If so, clear it and set it again.
  87          */
  88
  89         limit = 0;
  90
  91         inv_bit = dart_is_u4 ? DART_CNTL_U4_FLUSHTLB : DART_CNTL_U3_FLUSHTLB;
  92 retry:
  93         l = 0;
  94         reg = DART_IN(DART_CNTL);
  95         reg |= inv_bit;
  96         DART_OUT(DART_CNTL, reg);
  97
  98         while ((DART_IN(DART_CNTL) & inv_bit) && l < (1L << limit))
  99                 l++;
 100         if (l == (1L << limit)) {
 101                 if (limit < 4) {
 102                         limit++;
 103                         reg = DART_IN(DART_CNTL);
 104                         reg &= ~inv_bit;
 105                         DART_OUT(DART_CNTL, reg);
 106                         goto retry;
 107                 } else
 108                         panic("DART: TLB did not flush after waiting a long "
 109                               "time. Buggy U3 ?");
 110         }
 111 }
 112
 113 static void dart_flush(struct iommu_table *tbl)
 114 {
 115         if (dart_dirty)
 116                 dart_tlb_invalidate_all();
 117         dart_dirty = 0;
 118 }
 119
 120 static void dart_build(struct iommu_table *tbl, long index,
 121                        long npages, unsigned long uaddr,
 122                        enum dma_data_direction direction)
 123 {
 124         unsigned int *dp;
 125         unsigned int rpn;
 126
 127         DBG("dart: build at: %lx, %lx, addr: %x\n", index, npages, uaddr);
 128
 129         index <<= DART_PAGE_FACTOR;
 130         npages <<= DART_PAGE_FACTOR;
 131
 132         dp = ((unsigned int*)tbl->it_base) + index;
 133
 134         /* On U3, all memory is contigous, so we can move this
 135          * out of the loop.
 136          */
 137         while (npages--) {
 138                 rpn = virt_to_abs(uaddr) >> DART_PAGE_SHIFT;
 139
 140                 *(dp++) = DARTMAP_VALID | (rpn & DARTMAP_RPNMASK);
 141
 142                 uaddr += DART_PAGE_SIZE;
 143         }
 144
 145         dart_dirty = 1;
 146 }
 147
 148
 149 static void dart_free(struct iommu_table *tbl, long index, long npages)
 150 {
 151         unsigned int *dp;
 152
 153         /* We don't worry about flushing the TLB cache. The only drawback of
 154          * not doing it is that we won't catch buggy device drivers doing
 155          * bad DMAs, but then no 32-bit architecture ever does either.
 156          */
 157
 158         DBG("dart: free at: %lx, %lx\n", index, npages);
 159
 160         index <<= DART_PAGE_FACTOR;
 161         npages <<= DART_PAGE_FACTOR;
 162
 163         dp  = ((unsigned int *)tbl->it_base) + index;
 164
 165         while (npages--)
 166                 *(dp++) = dart_emptyval;
 167 }
 168
 169
 170 static int dart_init(struct device_node *dart_node)
 171 {
 172         unsigned int i;
 173         unsigned long tmp, base, size;
 174         struct resource r;
 175
 176         if (dart_tablebase == 0 || dart_tablesize == 0) {
 177                 printk(KERN_INFO "DART: table not allocated, using "
 178                        "direct DMA\n");
 179                 return -ENODEV;
 180         }
 181
 182         if (of_address_to_resource(dart_node, 0, &r))
 183                 panic("DART: can't get register base ! ");
 184
 185         /* Make sure nothing from the DART range remains in the CPU cache
 186          * from a previous mapping that existed before the kernel took
 187          * over
 188          */
 189         flush_dcache_phys_range(dart_tablebase,
 190                                 dart_tablebase + dart_tablesize);
 191
 192         /* Allocate a spare page to map all invalid DART pages. We need to do
 193          * that to work around what looks like a problem with the HT bridge
 194          * prefetching into invalid pages and corrupting data
 195          */
 196         tmp = lmb_alloc(DART_PAGE_SIZE, DART_PAGE_SIZE);
 197         dart_emptyval = DARTMAP_VALID | ((tmp >> DART_PAGE_SHIFT) &
 198                                          DARTMAP_RPNMASK);
 199
 200         /* Map in DART registers */
 201         dart = ioremap(r.start, r.end - r.start + 1);
 202         if (dart == NULL)
 203                 panic("DART: Cannot map registers!");
 204
 205         /* Map in DART table */
 206         dart_vbase = ioremap(virt_to_abs(dart_tablebase), dart_tablesize);
 207
 208         /* Fill initial table */
 209         for (i = 0; i < dart_tablesize/4; i++)
 210                 dart_vbase[i] = dart_emptyval;
 211
 212         /* Initialize DART with table base and enable it. */
 213         base = dart_tablebase >> DART_PAGE_SHIFT;
 214         size = dart_tablesize >> DART_PAGE_SHIFT;
 215         if (dart_is_u4) {
 216                 size &= DART_SIZE_U4_SIZE_MASK;
 217                 DART_OUT(DART_BASE_U4, base);
 218                 DART_OUT(DART_SIZE_U4, size);
 219                 DART_OUT(DART_CNTL, DART_CNTL_U4_ENABLE);
 220         } else {
 221                 size &= DART_CNTL_U3_SIZE_MASK;
 222                 DART_OUT(DART_CNTL,
 223                          DART_CNTL_U3_ENABLE |
 224                          (base << DART_CNTL_U3_BASE_SHIFT) |
 225                          (size << DART_CNTL_U3_SIZE_SHIFT));
 226         }
 227
 228         /* Invalidate DART to get rid of possible stale TLBs */
 229         dart_tlb_invalidate_all();
 230
 231         printk(KERN_INFO "DART IOMMU initialized for %s type chipset\n",
 232                dart_is_u4 ? "U4" : "U3");
 233
 234         return 0;
 235 }
 236
 237 static void iommu_table_dart_setup(void)
 238 {
 239         iommu_table_dart.it_busno = 0;
 240         iommu_table_dart.it_offset = 0;
 241         /* it_size is in number of entries */
 242         iommu_table_dart.it_size = (dart_tablesize / sizeof(u32)) >> DART_PAGE_FACTOR;
 243
 244         /* Initialize the common IOMMU code */
 245         iommu_table_dart.it_base = (unsigned long)dart_vbase;
 246         iommu_table_dart.it_index = 0;
 247         iommu_table_dart.it_blocksize = 1;
 248         iommu_init_table(&iommu_table_dart);
 249
 250         /* Reserve the last page of the DART to avoid possible prefetch
 251          * past the DART mapped area
 252          */
 253         set_bit(iommu_table_dart.it_size - 1, iommu_table_dart.it_map);
 254 }
 255
 256 static void iommu_dev_setup_dart(struct pci_dev *dev)
 257 {
 258         struct device_node *dn;
 259
 260         /* We only have one iommu table on the mac for now, which makes
 261          * things simple. Setup all PCI devices to point to this table
 262          *
 263          * We must use pci_device_to_OF_node() to make sure that
 264          * we get the real "final" pointer to the device in the
 265          * pci_dev sysdata and not the temporary PHB one
 266          */
 267         dn = pci_device_to_OF_node(dev);
 268
 269         if (dn)
 270                 PCI_DN(dn)->iommu_table = &iommu_table_dart;
 271 }
 272
 273 static void iommu_bus_setup_dart(struct pci_bus *bus)
 274 {
 275         struct device_node *dn;
 276
 277         if (!iommu_table_dart_inited) {
 278                 iommu_table_dart_inited = 1;
 279                 iommu_table_dart_setup();
 280         }
 281
 282         dn = pci_bus_to_OF_node(bus);
 283
 284         if (dn)
 285                 PCI_DN(dn)->iommu_table = &iommu_table_dart;
 286 }
 287
 288 static void iommu_dev_setup_null(struct pci_dev *dev) { }
 289 static void iommu_bus_setup_null(struct pci_bus *bus) { }
 290
 291 void iommu_init_early_dart(void)
 292 {
 293         struct device_node *dn;
 294
 295         /* Find the DART in the device-tree */
 296         dn = of_find_compatible_node(NULL, "dart", "u3-dart");
 297         if (dn == NULL) {
 298                 dn = of_find_compatible_node(NULL, "dart", "u4-dart");
 299                 if (dn == NULL)
 300                         goto bail;
 301                 dart_is_u4 = 1;
 302         }
 303
 304         /* Setup low level TCE operations for the core IOMMU code */
 305         ppc_md.tce_build = dart_build;
 306         ppc_md.tce_free  = dart_free;
 307         ppc_md.tce_flush = dart_flush;
 308
 309         /* Initialize the DART HW */
 310         if (dart_init(dn) == 0) {
 311                 ppc_md.iommu_dev_setup = iommu_dev_setup_dart;
 312                 ppc_md.iommu_bus_setup = iommu_bus_setup_dart;
 313
 314                 /* Setup pci_dma ops */
 315                 pci_iommu_init();
 316
 317                 return;
 318         }
 319
 320  bail:
 321         /* If init failed, use direct iommu and null setup functions */
 322         ppc_md.iommu_dev_setup = iommu_dev_setup_null;
 323         ppc_md.iommu_bus_setup = iommu_bus_setup_null;
 324
 325         /* Setup pci_dma ops */
 326         pci_direct_iommu_init();
 327 }
 328
 329
 330 void __init alloc_dart_table(void)
 331 {
 332         /* Only reserve DART space if machine has more than 2GB of RAM
 333          * or if requested with iommu=on on cmdline.
 334          */
 335         if (lmb_end_of_DRAM() <= 0x80000000ull && !iommu_force_on)
 336                 return;
 337
 338         /* 512 pages (2MB) is max DART tablesize. */
 339         dart_tablesize = 1UL << 21;
 340         /* 16MB (1 << 24) alignment. We allocate a full 16Mb chuck since we
 341          * will blow up an entire large page anyway in the kernel mapping
 342          */
 343         dart_tablebase = (unsigned long)
 344                 abs_to_virt(lmb_alloc_base(1UL<<24, 1UL<<24, 0x80000000L));
 345
 346         printk(KERN_INFO "DART table allocated at: %lx\n", dart_tablebase);
 347 }