drivers/iommu/irq_remapping.c

   1 #include <linux/seq_file.h>
   2 #include <linux/cpumask.h>
   3 #include <linux/kernel.h>
   4 #include <linux/string.h>
   5 #include <linux/errno.h>
   6 #include <linux/msi.h>
   7 #include <linux/irq.h>
   8 #include <linux/pci.h>
   9
  10 #include <asm/hw_irq.h>
  11 #include <asm/irq_remapping.h>
  12 #include <asm/processor.h>
  13 #include <asm/x86_init.h>
  14 #include <asm/apic.h>
  15 #include <asm/hpet.h>
  16
  17 #include "irq_remapping.h"
  18
  19 int irq_remapping_enabled;
  20
  21 int disable_irq_remap;
  22 int irq_remap_broken;
  23 int disable_sourceid_checking;
  24 int no_x2apic_optout;
  25
  26 static struct irq_remap_ops *remap_ops;
  27
  28 static int msi_alloc_remapped_irq(struct pci_dev *pdev, int irq, int nvec);
  29 static int msi_setup_remapped_irq(struct pci_dev *pdev, unsigned int irq,
  30                                   int index, int sub_handle);
  31 static int set_remapped_irq_affinity(struct irq_data *data,
  32                                      const struct cpumask *mask,
  33                                      bool force);
  34
  35 static bool irq_remapped(struct irq_cfg *cfg)
  36 {
  37         return (cfg->remapped == 1);
  38 }
  39
  40 static void irq_remapping_disable_io_apic(void)
  41 {
  42         /*
  43          * With interrupt-remapping, for now we will use virtual wire A
  44          * mode, as virtual wire B is little complex (need to configure
  45          * both IOAPIC RTE as well as interrupt-remapping table entry).
  46          * As this gets called during crash dump, keep this simple for
  47          * now.
  48          */
  49         if (cpu_has_apic || apic_from_smp_config())
  50                 disconnect_bsp_APIC(0);
  51 }
  52
  53 static int do_setup_msi_irqs(struct pci_dev *dev, int nvec)
  54 {
  55         int ret, sub_handle, nvec_pow2, index = 0;
  56         unsigned int irq;
  57         struct msi_desc *msidesc;
  58
  59         WARN_ON(!list_is_singular(&dev->msi_list));
  60         msidesc = list_entry(dev->msi_list.next, struct msi_desc, list);
  61         WARN_ON(msidesc->irq);
  62         WARN_ON(msidesc->msi_attrib.multiple);
  63         WARN_ON(msidesc->nvec_used);
  64
  65         irq = irq_alloc_hwirqs(nvec, dev_to_node(&dev->dev));
  66         if (irq == 0)
  67                 return -ENOSPC;
  68
  69         nvec_pow2 = __roundup_pow_of_two(nvec);
  70         msidesc->nvec_used = nvec;
  71         msidesc->msi_attrib.multiple = ilog2(nvec_pow2);
  72         for (sub_handle = 0; sub_handle < nvec; sub_handle++) {
  73                 if (!sub_handle) {
  74                         index = msi_alloc_remapped_irq(dev, irq, nvec_pow2);
  75                         if (index < 0) {
  76                                 ret = index;
  77                                 goto error;
  78                         }
  79                 } else {
  80                         ret = msi_setup_remapped_irq(dev, irq + sub_handle,
  81                                                      index, sub_handle);
  82                         if (ret < 0)
  83                                 goto error;
  84                 }
  85                 ret = setup_msi_irq(dev, msidesc, irq, sub_handle);
  86                 if (ret < 0)
  87                         goto error;
  88         }
  89         return 0;
  90
  91 error:
  92         irq_free_hwirqs(irq, nvec);
  93
  94         /*
  95          * Restore altered MSI descriptor fields and prevent just destroyed
  96          * IRQs from tearing down again in default_teardown_msi_irqs()
  97          */
  98         msidesc->irq = 0;
  99         msidesc->nvec_used = 0;
 100         msidesc->msi_attrib.multiple = 0;
 101
 102         return ret;
 103 }
 104
 105 static int do_setup_msix_irqs(struct pci_dev *dev, int nvec)
 106 {
 107         int node, ret, sub_handle, index = 0;
 108         struct msi_desc *msidesc;
 109         unsigned int irq;
 110
 111         node            = dev_to_node(&dev->dev);
 112         sub_handle      = 0;
 113
 114         list_for_each_entry(msidesc, &dev->msi_list, list) {
 115
 116                 irq = irq_alloc_hwirq(node);
 117                 if (irq == 0)
 118                         return -1;
 119
 120                 if (sub_handle == 0)
 121                         ret = index = msi_alloc_remapped_irq(dev, irq, nvec);
 122                 else
 123                         ret = msi_setup_remapped_irq(dev, irq, index, sub_handle);
 124
 125                 if (ret < 0)
 126                         goto error;
 127
 128                 ret = setup_msi_irq(dev, msidesc, irq, 0);
 129                 if (ret < 0)
 130                         goto error;
 131
 132                 sub_handle += 1;
 133                 irq        += 1;
 134         }
 135
 136         return 0;
 137
 138 error:
 139         irq_free_hwirq(irq);
 140         return ret;
 141 }
 142
 143 static int irq_remapping_setup_msi_irqs(struct pci_dev *dev,
 144                                         int nvec, int type)
 145 {
 146         if (type == PCI_CAP_ID_MSI)
 147                 return do_setup_msi_irqs(dev, nvec);
 148         else
 149                 return do_setup_msix_irqs(dev, nvec);
 150 }
 151
 152 static void eoi_ioapic_pin_remapped(int apic, int pin, int vector)
 153 {
 154         /*
 155          * Intr-remapping uses pin number as the virtual vector
 156          * in the RTE. Actual vector is programmed in
 157          * intr-remapping table entry. Hence for the io-apic
 158          * EOI we use the pin number.
 159          */
 160         io_apic_eoi(apic, pin);
 161 }
 162
 163 static void __init irq_remapping_modify_x86_ops(void)
 164 {
 165         x86_io_apic_ops.disable         = irq_remapping_disable_io_apic;
 166         x86_io_apic_ops.set_affinity    = set_remapped_irq_affinity;
 167         x86_io_apic_ops.setup_entry     = setup_ioapic_remapped_entry;
 168         x86_io_apic_ops.eoi_ioapic_pin  = eoi_ioapic_pin_remapped;
 169         x86_msi.setup_msi_irqs          = irq_remapping_setup_msi_irqs;
 170         x86_msi.setup_hpet_msi          = setup_hpet_msi_remapped;
 171         x86_msi.compose_msi_msg         = compose_remapped_msi_msg;
 172 }
 173
 174 static __init int setup_nointremap(char *str)
 175 {
 176         disable_irq_remap = 1;
 177         return 0;
 178 }
 179 early_param("nointremap", setup_nointremap);
 180
 181 static __init int setup_irqremap(char *str)
 182 {
 183         if (!str)
 184                 return -EINVAL;
 185
 186         while (*str) {
 187                 if (!strncmp(str, "on", 2))
 188                         disable_irq_remap = 0;
 189                 else if (!strncmp(str, "off", 3))
 190                         disable_irq_remap = 1;
 191                 else if (!strncmp(str, "nosid", 5))
 192                         disable_sourceid_checking = 1;
 193                 else if (!strncmp(str, "no_x2apic_optout", 16))
 194                         no_x2apic_optout = 1;
 195
 196                 str += strcspn(str, ",");
 197                 while (*str == ',')
 198                         str++;
 199         }
 200
 201         return 0;
 202 }
 203 early_param("intremap", setup_irqremap);
 204
 205 void __init setup_irq_remapping_ops(void)
 206 {
 207         remap_ops = &intel_irq_remap_ops;
 208
 209 #ifdef CONFIG_AMD_IOMMU
 210         if (amd_iommu_irq_ops.prepare() == 0)
 211                 remap_ops = &amd_iommu_irq_ops;
 212 #endif
 213 }
 214
 215 void set_irq_remapping_broken(void)
 216 {
 217         irq_remap_broken = 1;
 218 }
 219
 220 int irq_remapping_supported(void)
 221 {
 222         if (disable_irq_remap)
 223                 return 0;
 224
 225         if (!remap_ops || !remap_ops->supported)
 226                 return 0;
 227
 228         return remap_ops->supported();
 229 }
 230
 231 int __init irq_remapping_prepare(void)
 232 {
 233         if (!remap_ops || !remap_ops->prepare)
 234                 return -ENODEV;
 235
 236         return remap_ops->prepare();
 237 }
 238
 239 int __init irq_remapping_enable(void)
 240 {
 241         int ret;
 242
 243         if (!remap_ops || !remap_ops->enable)
 244                 return -ENODEV;
 245
 246         ret = remap_ops->enable();
 247
 248         if (irq_remapping_enabled)
 249                 irq_remapping_modify_x86_ops();
 250
 251         return ret;
 252 }
 253
 254 void irq_remapping_disable(void)
 255 {
 256         if (!irq_remapping_enabled ||
 257             !remap_ops ||
 258             !remap_ops->disable)
 259                 return;
 260
 261         remap_ops->disable();
 262 }
 263
 264 int irq_remapping_reenable(int mode)
 265 {
 266         if (!irq_remapping_enabled ||
 267             !remap_ops ||
 268             !remap_ops->reenable)
 269                 return 0;
 270
 271         return remap_ops->reenable(mode);
 272 }
 273
 274 int __init irq_remap_enable_fault_handling(void)
 275 {
 276         if (!irq_remapping_enabled)
 277                 return 0;
 278
 279         if (!remap_ops || !remap_ops->enable_faulting)
 280                 return -ENODEV;
 281
 282         return remap_ops->enable_faulting();
 283 }
 284
 285 int setup_ioapic_remapped_entry(int irq,
 286                                 struct IO_APIC_route_entry *entry,
 287                                 unsigned int destination, int vector,
 288                                 struct io_apic_irq_attr *attr)
 289 {
 290         if (!remap_ops || !remap_ops->setup_ioapic_entry)
 291                 return -ENODEV;
 292
 293         return remap_ops->setup_ioapic_entry(irq, entry, destination,
 294                                              vector, attr);
 295 }
 296
 297 static int set_remapped_irq_affinity(struct irq_data *data,
 298                                      const struct cpumask *mask, bool force)
 299 {
 300         if (!config_enabled(CONFIG_SMP) || !remap_ops ||
 301             !remap_ops->set_affinity)
 302                 return 0;
 303
 304         return remap_ops->set_affinity(data, mask, force);
 305 }
 306
 307 void free_remapped_irq(int irq)
 308 {
 309         struct irq_cfg *cfg = irq_get_chip_data(irq);
 310
 311         if (!remap_ops || !remap_ops->free_irq)
 312                 return;
 313
 314         if (irq_remapped(cfg))
 315                 remap_ops->free_irq(irq);
 316 }
 317
 318 void compose_remapped_msi_msg(struct pci_dev *pdev,
 319                               unsigned int irq, unsigned int dest,
 320                               struct msi_msg *msg, u8 hpet_id)
 321 {
 322         struct irq_cfg *cfg = irq_get_chip_data(irq);
 323
 324         if (!irq_remapped(cfg))
 325                 native_compose_msi_msg(pdev, irq, dest, msg, hpet_id);
 326         else if (remap_ops && remap_ops->compose_msi_msg)
 327                 remap_ops->compose_msi_msg(pdev, irq, dest, msg, hpet_id);
 328 }
 329
 330 static int msi_alloc_remapped_irq(struct pci_dev *pdev, int irq, int nvec)
 331 {
 332         if (!remap_ops || !remap_ops->msi_alloc_irq)
 333                 return -ENODEV;
 334
 335         return remap_ops->msi_alloc_irq(pdev, irq, nvec);
 336 }
 337
 338 static int msi_setup_remapped_irq(struct pci_dev *pdev, unsigned int irq,
 339                                   int index, int sub_handle)
 340 {
 341         if (!remap_ops || !remap_ops->msi_setup_irq)
 342                 return -ENODEV;
 343
 344         return remap_ops->msi_setup_irq(pdev, irq, index, sub_handle);
 345 }
 346
 347 int setup_hpet_msi_remapped(unsigned int irq, unsigned int id)
 348 {
 349         int ret;
 350
 351         if (!remap_ops || !remap_ops->alloc_hpet_msi)
 352                 return -ENODEV;
 353
 354         ret = remap_ops->alloc_hpet_msi(irq, id);
 355         if (ret)
 356                 return -EINVAL;
 357
 358         return default_setup_hpet_msi(irq, id);
 359 }
 360
 361 void panic_if_irq_remap(const char *msg)
 362 {
 363         if (irq_remapping_enabled)
 364                 panic(msg);
 365 }
 366
 367 static void ir_ack_apic_edge(struct irq_data *data)
 368 {
 369         ack_APIC_irq();
 370 }
 371
 372 static void ir_ack_apic_level(struct irq_data *data)
 373 {
 374         ack_APIC_irq();
 375         eoi_ioapic_irq(data->irq, data->chip_data);
 376 }
 377
 378 static void ir_print_prefix(struct irq_data *data, struct seq_file *p)
 379 {
 380         seq_printf(p, " IR-%s", data->chip->name);
 381 }
 382
 383 void irq_remap_modify_chip_defaults(struct irq_chip *chip)
 384 {
 385         chip->irq_print_chip = ir_print_prefix;
 386         chip->irq_ack = ir_ack_apic_edge;
 387         chip->irq_eoi = ir_ack_apic_level;
 388         chip->irq_set_affinity = x86_io_apic_ops.set_affinity;
 389 }
 390
 391 bool setup_remapped_irq(int irq, struct irq_cfg *cfg, struct irq_chip *chip)
 392 {
 393         if (!irq_remapped(cfg))
 394                 return false;
 395         irq_set_status_flags(irq, IRQ_MOVE_PCNTXT);
 396         irq_remap_modify_chip_defaults(chip);
 397         return true;
 398 }