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