arch/x86/kernel/pci-dma.c

   1 #include <linux/dma-mapping.h>
   2 #include <linux/dma-debug.h>
   3 #include <linux/dmar.h>
   4 #include <linux/export.h>
   5 #include <linux/bootmem.h>
   6 #include <linux/gfp.h>
   7 #include <linux/pci.h>
   8 #include <linux/kmemleak.h>
   9
  10 #include <asm/proto.h>
  11 #include <asm/dma.h>
  12 #include <asm/iommu.h>
  13 #include <asm/gart.h>
  14 #include <asm/calgary.h>
  15 #include <asm/x86_init.h>
  16 #include <asm/iommu_table.h>
  17
  18 static int forbid_dac __read_mostly;
  19
  20 struct dma_map_ops *dma_ops = &nommu_dma_ops;
  21 EXPORT_SYMBOL(dma_ops);
  22
  23 static int iommu_sac_force __read_mostly;
  24
  25 #ifdef CONFIG_IOMMU_DEBUG
  26 int panic_on_overflow __read_mostly = 1;
  27 int force_iommu __read_mostly = 1;
  28 #else
  29 int panic_on_overflow __read_mostly = 0;
  30 int force_iommu __read_mostly = 0;
  31 #endif
  32
  33 int iommu_merge __read_mostly = 0;
  34
  35 int no_iommu __read_mostly;
  36 /* Set this to 1 if there is a HW IOMMU in the system */
  37 int iommu_detected __read_mostly = 0;
  38
  39 /*
  40  * This variable becomes 1 if iommu=pt is passed on the kernel command line.
  41  * If this variable is 1, IOMMU implementations do no DMA translation for
  42  * devices and allow every device to access to whole physical memory. This is
  43  * useful if a user wants to use an IOMMU only for KVM device assignment to
  44  * guests and not for driver dma translation.
  45  */
  46 int iommu_pass_through __read_mostly;
  47
  48 extern struct iommu_table_entry __iommu_table[], __iommu_table_end[];
  49
  50 /* Dummy device used for NULL arguments (normally ISA). */
  51 struct device x86_dma_fallback_dev = {
  52         .init_name = "fallback device",
  53         .coherent_dma_mask = ISA_DMA_BIT_MASK,
  54         .dma_mask = &x86_dma_fallback_dev.coherent_dma_mask,
  55 };
  56 EXPORT_SYMBOL(x86_dma_fallback_dev);
  57
  58 /* Number of entries preallocated for DMA-API debugging */
  59 #define PREALLOC_DMA_DEBUG_ENTRIES       65536
  60
  61 void __init pci_iommu_alloc(void)
  62 {
  63         struct iommu_table_entry *p;
  64
  65         sort_iommu_table(__iommu_table, __iommu_table_end);
  66         check_iommu_entries(__iommu_table, __iommu_table_end);
  67
  68         for (p = __iommu_table; p < __iommu_table_end; p++) {
  69                 if (p && p->detect && p->detect() > 0) {
  70                         p->flags |= IOMMU_DETECTED;
  71                         if (p->early_init)
  72                                 p->early_init();
  73                         if (p->flags & IOMMU_FINISH_IF_DETECTED)
  74                                 break;
  75                 }
  76         }
  77 }
  78 void *dma_generic_alloc_coherent(struct device *dev, size_t size,
  79                                  dma_addr_t *dma_addr, gfp_t flag,
  80                                  struct dma_attrs *attrs)
  81 {
  82         unsigned long dma_mask;
  83         struct page *page;
  84         unsigned int count = PAGE_ALIGN(size) >> PAGE_SHIFT;
  85         dma_addr_t addr;
  86
  87         dma_mask = dma_alloc_coherent_mask(dev, flag);
  88
  89         flag &= ~__GFP_ZERO;
  90 again:
  91         page = NULL;
  92         /* CMA can be used only in the context which permits sleeping */
  93         if (gfpflags_allow_blocking(flag)) {
  94                 page = dma_alloc_from_contiguous(dev, count, get_order(size));
  95                 if (page && page_to_phys(page) + size > dma_mask) {
  96                         dma_release_from_contiguous(dev, page, count);
  97                         page = NULL;
  98                 }
  99         }
 100         /* fallback */
 101         if (!page)
 102                 page = alloc_pages_node(dev_to_node(dev), flag, get_order(size));
 103         if (!page)
 104                 return NULL;
 105
 106         addr = page_to_phys(page);
 107         if (addr + size > dma_mask) {
 108                 __free_pages(page, get_order(size));
 109
 110                 if (dma_mask < DMA_BIT_MASK(32) && !(flag & GFP_DMA)) {
 111                         flag = (flag & ~GFP_DMA32) | GFP_DMA;
 112                         goto again;
 113                 }
 114
 115                 return NULL;
 116         }
 117         memset(page_address(page), 0, size);
 118         *dma_addr = addr;
 119         return page_address(page);
 120 }
 121
 122 void dma_generic_free_coherent(struct device *dev, size_t size, void *vaddr,
 123                                dma_addr_t dma_addr, struct dma_attrs *attrs)
 124 {
 125         unsigned int count = PAGE_ALIGN(size) >> PAGE_SHIFT;
 126         struct page *page = virt_to_page(vaddr);
 127
 128         if (!dma_release_from_contiguous(dev, page, count))
 129                 free_pages((unsigned long)vaddr, get_order(size));
 130 }
 131
 132 bool arch_dma_alloc_attrs(struct device **dev, gfp_t *gfp)
 133 {
 134         if (!*dev)
 135                 *dev = &x86_dma_fallback_dev;
 136
 137         *gfp &= ~(__GFP_DMA | __GFP_HIGHMEM | __GFP_DMA32);
 138         *gfp = dma_alloc_coherent_gfp_flags(*dev, *gfp);
 139
 140         if (!is_device_dma_capable(*dev))
 141                 return false;
 142         return true;
 143
 144 }
 145 EXPORT_SYMBOL(arch_dma_alloc_attrs);
 146
 147 /*
 148  * See <Documentation/x86/x86_64/boot-options.txt> for the iommu kernel
 149  * parameter documentation.
 150  */
 151 static __init int iommu_setup(char *p)
 152 {
 153         iommu_merge = 1;
 154
 155         if (!p)
 156                 return -EINVAL;
 157
 158         while (*p) {
 159                 if (!strncmp(p, "off", 3))
 160                         no_iommu = 1;
 161                 /* gart_parse_options has more force support */
 162                 if (!strncmp(p, "force", 5))
 163                         force_iommu = 1;
 164                 if (!strncmp(p, "noforce", 7)) {
 165                         iommu_merge = 0;
 166                         force_iommu = 0;
 167                 }
 168
 169                 if (!strncmp(p, "biomerge", 8)) {
 170                         iommu_merge = 1;
 171                         force_iommu = 1;
 172                 }
 173                 if (!strncmp(p, "panic", 5))
 174                         panic_on_overflow = 1;
 175                 if (!strncmp(p, "nopanic", 7))
 176                         panic_on_overflow = 0;
 177                 if (!strncmp(p, "merge", 5)) {
 178                         iommu_merge = 1;
 179                         force_iommu = 1;
 180                 }
 181                 if (!strncmp(p, "nomerge", 7))
 182                         iommu_merge = 0;
 183                 if (!strncmp(p, "forcesac", 8))
 184                         iommu_sac_force = 1;
 185                 if (!strncmp(p, "allowdac", 8))
 186                         forbid_dac = 0;
 187                 if (!strncmp(p, "nodac", 5))
 188                         forbid_dac = 1;
 189                 if (!strncmp(p, "usedac", 6)) {
 190                         forbid_dac = -1;
 191                         return 1;
 192                 }
 193 #ifdef CONFIG_SWIOTLB
 194                 if (!strncmp(p, "soft", 4))
 195                         swiotlb = 1;
 196 #endif
 197                 if (!strncmp(p, "pt", 2))
 198                         iommu_pass_through = 1;
 199
 200                 gart_parse_options(p);
 201
 202 #ifdef CONFIG_CALGARY_IOMMU
 203                 if (!strncmp(p, "calgary", 7))
 204                         use_calgary = 1;
 205 #endif /* CONFIG_CALGARY_IOMMU */
 206
 207                 p += strcspn(p, ",");
 208                 if (*p == ',')
 209                         ++p;
 210         }
 211         return 0;
 212 }
 213 early_param("iommu", iommu_setup);
 214
 215 int dma_supported(struct device *dev, u64 mask)
 216 {
 217         struct dma_map_ops *ops = get_dma_ops(dev);
 218
 219 #ifdef CONFIG_PCI
 220         if (mask > 0xffffffff && forbid_dac > 0) {
 221                 dev_info(dev, "PCI: Disallowing DAC for device\n");
 222                 return 0;
 223         }
 224 #endif
 225
 226         if (ops->dma_supported)
 227                 return ops->dma_supported(dev, mask);
 228
 229         /* Copied from i386. Doesn't make much sense, because it will
 230            only work for pci_alloc_coherent.
 231            The caller just has to use GFP_DMA in this case. */
 232         if (mask < DMA_BIT_MASK(24))
 233                 return 0;
 234
 235         /* Tell the device to use SAC when IOMMU force is on.  This
 236            allows the driver to use cheaper accesses in some cases.
 237
 238            Problem with this is that if we overflow the IOMMU area and
 239            return DAC as fallback address the device may not handle it
 240            correctly.
 241
 242            As a special case some controllers have a 39bit address
 243            mode that is as efficient as 32bit (aic79xx). Don't force
 244            SAC for these.  Assume all masks <= 40 bits are of this
 245            type. Normally this doesn't make any difference, but gives
 246            more gentle handling of IOMMU overflow. */
 247         if (iommu_sac_force && (mask >= DMA_BIT_MASK(40))) {
 248                 dev_info(dev, "Force SAC with mask %Lx\n", mask);
 249                 return 0;
 250         }
 251
 252         return 1;
 253 }
 254 EXPORT_SYMBOL(dma_supported);
 255
 256 static int __init pci_iommu_init(void)
 257 {
 258         struct iommu_table_entry *p;
 259         dma_debug_init(PREALLOC_DMA_DEBUG_ENTRIES);
 260
 261 #ifdef CONFIG_PCI
 262         dma_debug_add_bus(&pci_bus_type);
 263 #endif
 264         x86_init.iommu.iommu_init();
 265
 266         for (p = __iommu_table; p < __iommu_table_end; p++) {
 267                 if (p && (p->flags & IOMMU_DETECTED) && p->late_init)
 268                         p->late_init();
 269         }
 270
 271         return 0;
 272 }
 273 /* Must execute after PCI subsystem */
 274 rootfs_initcall(pci_iommu_init);
 275
 276 #ifdef CONFIG_PCI
 277 /* Many VIA bridges seem to corrupt data for DAC. Disable it here */
 278
 279 static void via_no_dac(struct pci_dev *dev)
 280 {
 281         if (forbid_dac == 0) {
 282                 dev_info(&dev->dev, "disabling DAC on VIA PCI bridge\n");
 283                 forbid_dac = 1;
 284         }
 285 }
 286 DECLARE_PCI_FIXUP_CLASS_FINAL(PCI_VENDOR_ID_VIA, PCI_ANY_ID,
 287                                 PCI_CLASS_BRIDGE_PCI, 8, via_no_dac);
 288 #endif