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 16 #include "irq_remapping.h" 17 18 int irq_remapping_enabled; 19 20 int disable_irq_remap; 21 int irq_remap_broken; 22 int disable_sourceid_checking; 23 int no_x2apic_optout; 24 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 node, ret, sub_handle, nvec_pow2, index = 0; 55 unsigned int irq; 56 struct msi_desc *msidesc; 57 58 WARN_ON(!list_is_singular(&dev->msi_list)); 59 msidesc = list_entry(dev->msi_list.next, struct msi_desc, list); 60 WARN_ON(msidesc->irq); 61 WARN_ON(msidesc->msi_attrib.multiple); 62 WARN_ON(msidesc->nvec_used); 63 64 node = dev_to_node(&dev->dev); 65 irq = __create_irqs(get_nr_irqs_gsi(), nvec, node); 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 destroy_irqs(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 irq = get_nr_irqs_gsi(); 113 sub_handle = 0; 114 115 list_for_each_entry(msidesc, &dev->msi_list, list) { 116 117 irq = create_irq_nr(irq, node); 118 if (irq == 0) 119 return -1; 120 121 if (sub_handle == 0) 122 ret = index = msi_alloc_remapped_irq(dev, irq, nvec); 123 else 124 ret = msi_setup_remapped_irq(dev, irq, index, sub_handle); 125 126 if (ret < 0) 127 goto error; 128 129 ret = setup_msi_irq(dev, msidesc, irq, 0); 130 if (ret < 0) 131 goto error; 132 133 sub_handle += 1; 134 irq += 1; 135 } 136 137 return 0; 138 139 error: 140 destroy_irq(irq); 141 return ret; 142 } 143 144 static int irq_remapping_setup_msi_irqs(struct pci_dev *dev, 145 int nvec, int type) 146 { 147 if (type == PCI_CAP_ID_MSI) 148 return do_setup_msi_irqs(dev, nvec); 149 else 150 return do_setup_msix_irqs(dev, nvec); 151 } 152 153 void eoi_ioapic_pin_remapped(int apic, int pin, int vector) 154 { 155 /* 156 * Intr-remapping uses pin number as the virtual vector 157 * in the RTE. Actual vector is programmed in 158 * intr-remapping table entry. Hence for the io-apic 159 * EOI we use the pin number. 160 */ 161 io_apic_eoi(apic, pin); 162 } 163 164 static void __init irq_remapping_modify_x86_ops(void) 165 { 166 x86_io_apic_ops.disable = irq_remapping_disable_io_apic; 167 x86_io_apic_ops.set_affinity = set_remapped_irq_affinity; 168 x86_io_apic_ops.setup_entry = setup_ioapic_remapped_entry; 169 x86_io_apic_ops.eoi_ioapic_pin = eoi_ioapic_pin_remapped; 170 x86_msi.setup_msi_irqs = irq_remapping_setup_msi_irqs; 171 x86_msi.setup_hpet_msi = setup_hpet_msi_remapped; 172 x86_msi.compose_msi_msg = compose_remapped_msi_msg; 173 } 174 175 static __init int setup_nointremap(char *str) 176 { 177 disable_irq_remap = 1; 178 return 0; 179 } 180 early_param("nointremap", setup_nointremap); 181 182 static __init int setup_irqremap(char *str) 183 { 184 if (!str) 185 return -EINVAL; 186 187 while (*str) { 188 if (!strncmp(str, "on", 2)) 189 disable_irq_remap = 0; 190 else if (!strncmp(str, "off", 3)) 191 disable_irq_remap = 1; 192 else if (!strncmp(str, "nosid", 5)) 193 disable_sourceid_checking = 1; 194 else if (!strncmp(str, "no_x2apic_optout", 16)) 195 no_x2apic_optout = 1; 196 197 str += strcspn(str, ","); 198 while (*str == ',') 199 str++; 200 } 201 202 return 0; 203 } 204 early_param("intremap", setup_irqremap); 205 206 void __init setup_irq_remapping_ops(void) 207 { 208 remap_ops = &intel_irq_remap_ops; 209 210 #ifdef CONFIG_AMD_IOMMU 211 if (amd_iommu_irq_ops.prepare() == 0) 212 remap_ops = &amd_iommu_irq_ops; 213 #endif 214 } 215 216 void set_irq_remapping_broken(void) 217 { 218 irq_remap_broken = 1; 219 } 220 221 int irq_remapping_supported(void) 222 { 223 if (disable_irq_remap) 224 return 0; 225 226 if (!remap_ops || !remap_ops->supported) 227 return 0; 228 229 return remap_ops->supported(); 230 } 231 232 int __init irq_remapping_prepare(void) 233 { 234 if (!remap_ops || !remap_ops->prepare) 235 return -ENODEV; 236 237 return remap_ops->prepare(); 238 } 239 240 int __init irq_remapping_enable(void) 241 { 242 int ret; 243 244 if (!remap_ops || !remap_ops->enable) 245 return -ENODEV; 246 247 ret = remap_ops->enable(); 248 249 if (irq_remapping_enabled) 250 irq_remapping_modify_x86_ops(); 251 252 return ret; 253 } 254 255 void irq_remapping_disable(void) 256 { 257 if (!irq_remapping_enabled || 258 !remap_ops || 259 !remap_ops->disable) 260 return; 261 262 remap_ops->disable(); 263 } 264 265 int irq_remapping_reenable(int mode) 266 { 267 if (!irq_remapping_enabled || 268 !remap_ops || 269 !remap_ops->reenable) 270 return 0; 271 272 return remap_ops->reenable(mode); 273 } 274 275 int __init irq_remap_enable_fault_handling(void) 276 { 277 if (!irq_remapping_enabled) 278 return 0; 279 280 if (!remap_ops || !remap_ops->enable_faulting) 281 return -ENODEV; 282 283 return remap_ops->enable_faulting(); 284 } 285 286 int setup_ioapic_remapped_entry(int irq, 287 struct IO_APIC_route_entry *entry, 288 unsigned int destination, int vector, 289 struct io_apic_irq_attr *attr) 290 { 291 if (!remap_ops || !remap_ops->setup_ioapic_entry) 292 return -ENODEV; 293 294 return remap_ops->setup_ioapic_entry(irq, entry, destination, 295 vector, attr); 296 } 297 298 int set_remapped_irq_affinity(struct irq_data *data, const struct cpumask *mask, 299 bool force) 300 { 301 if (!config_enabled(CONFIG_SMP) || !remap_ops || 302 !remap_ops->set_affinity) 303 return 0; 304 305 return remap_ops->set_affinity(data, mask, force); 306 } 307 308 void free_remapped_irq(int irq) 309 { 310 struct irq_cfg *cfg = irq_get_chip_data(irq); 311 312 if (!remap_ops || !remap_ops->free_irq) 313 return; 314 315 if (irq_remapped(cfg)) 316 remap_ops->free_irq(irq); 317 } 318 319 void compose_remapped_msi_msg(struct pci_dev *pdev, 320 unsigned int irq, unsigned int dest, 321 struct msi_msg *msg, u8 hpet_id) 322 { 323 struct irq_cfg *cfg = irq_get_chip_data(irq); 324 325 if (!irq_remapped(cfg)) 326 native_compose_msi_msg(pdev, irq, dest, msg, hpet_id); 327 else if (remap_ops && remap_ops->compose_msi_msg) 328 remap_ops->compose_msi_msg(pdev, irq, dest, msg, hpet_id); 329 } 330 331 static int msi_alloc_remapped_irq(struct pci_dev *pdev, int irq, int nvec) 332 { 333 if (!remap_ops || !remap_ops->msi_alloc_irq) 334 return -ENODEV; 335 336 return remap_ops->msi_alloc_irq(pdev, irq, nvec); 337 } 338 339 static int msi_setup_remapped_irq(struct pci_dev *pdev, unsigned int irq, 340 int index, int sub_handle) 341 { 342 if (!remap_ops || !remap_ops->msi_setup_irq) 343 return -ENODEV; 344 345 return remap_ops->msi_setup_irq(pdev, irq, index, sub_handle); 346 } 347 348 int setup_hpet_msi_remapped(unsigned int irq, unsigned int id) 349 { 350 if (!remap_ops || !remap_ops->setup_hpet_msi) 351 return -ENODEV; 352 353 return remap_ops->setup_hpet_msi(irq, id); 354 } 355 356 void panic_if_irq_remap(const char *msg) 357 { 358 if (irq_remapping_enabled) 359 panic(msg); 360 } 361 362 static void ir_ack_apic_edge(struct irq_data *data) 363 { 364 ack_APIC_irq(); 365 } 366 367 static void ir_ack_apic_level(struct irq_data *data) 368 { 369 ack_APIC_irq(); 370 eoi_ioapic_irq(data->irq, data->chip_data); 371 } 372 373 static void ir_print_prefix(struct irq_data *data, struct seq_file *p) 374 { 375 seq_printf(p, " IR-%s", data->chip->name); 376 } 377 378 void irq_remap_modify_chip_defaults(struct irq_chip *chip) 379 { 380 chip->irq_print_chip = ir_print_prefix; 381 chip->irq_ack = ir_ack_apic_edge; 382 chip->irq_eoi = ir_ack_apic_level; 383 chip->irq_set_affinity = x86_io_apic_ops.set_affinity; 384 } 385 386 bool setup_remapped_irq(int irq, struct irq_cfg *cfg, struct irq_chip *chip) 387 { 388 if (!irq_remapped(cfg)) 389 return false; 390 irq_set_status_flags(irq, IRQ_MOVE_PCNTXT); 391 irq_remap_modify_chip_defaults(chip); 392 return true; 393 } 394